blob: 549bbe2f25c6df48e82873e72e61d46c00686174 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/personality.h>
25#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050026#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070033#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36
Eric Parise81e3f42009-12-04 15:47:36 -050037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800117static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
Al Virof52e0c12011-03-14 18:56:51 -0400139static char *getname_flags(const char __user * filename, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
Al Virof52e0c12011-03-14 18:56:51 -0400150 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 }
155 }
156 audit_getname(result);
157 return result;
158}
159
Al Virof52e0c12011-03-14 18:56:51 -0400160char *getname(const char __user * filename)
161{
162 return getname_flags(filename, 0);
163}
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165#ifdef CONFIG_AUDITSYSCALL
166void putname(const char *name)
167{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400168 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 audit_putname(name);
170 else
171 __putname(name);
172}
173EXPORT_SYMBOL(putname);
174#endif
175
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700176/*
177 * This does basic POSIX ACL permission checking
178 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100179static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
180 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700181{
182 umode_t mode = inode->i_mode;
183
184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
185
186 if (current_fsuid() == inode->i_uid)
187 mode >>= 6;
188 else {
189 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100190 int error = check_acl(inode, mask, flags);
191 if (error != -EAGAIN)
192 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700193 }
194
195 if (in_group_p(inode->i_gid))
196 mode >>= 3;
197 }
198
199 /*
200 * If the DACs are ok we don't need any capability check.
201 */
202 if ((mask & ~mode) == 0)
203 return 0;
204 return -EACCES;
205}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100208 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 * @inode: inode to check access rights for
210 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
211 * @check_acl: optional callback to check for Posix ACLs
Randy Dunlap39191622011-01-08 19:36:21 -0800212 * @flags: IPERM_FLAG_ flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 *
214 * Used to check for read/write/execute permissions on a file.
215 * We use "fsuid" for this, letting us set arbitrary permissions
216 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100217 * are used for other things.
218 *
219 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
220 * request cannot be satisfied (eg. requires blocking or too much complexity).
221 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100223int generic_permission(struct inode *inode, int mask, unsigned int flags,
224 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700226 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
228 /*
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700229 * Do the basic POSIX ACL permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100231 ret = acl_permission_check(inode, mask, flags, check_acl);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700232 if (ret != -EACCES)
233 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 /*
236 * Read/write DACs are always overridable.
237 * Executable DACs are overridable if at least one exec bit is set.
238 */
Miklos Szeredif696a362008-07-31 13:41:58 +0200239 if (!(mask & MAY_EXEC) || execute_ok(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 if (capable(CAP_DAC_OVERRIDE))
241 return 0;
242
243 /*
244 * Searching includes executable on directories, else just read.
245 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600246 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
248 if (capable(CAP_DAC_READ_SEARCH))
249 return 0;
250
251 return -EACCES;
252}
253
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200254/**
255 * inode_permission - check for access rights to a given inode
256 * @inode: inode to check permission on
257 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
258 *
259 * Used to check for read/write/execute permissions on an inode.
260 * We use "fsuid" for this, letting us set arbitrary permissions
261 * for filesystem access without changing the "normal" uids which
262 * are used for other things.
263 */
Al Virof419a2e2008-07-22 00:07:17 -0400264int inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
Al Viroe6305c42008-07-15 21:03:57 -0400266 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268 if (mask & MAY_WRITE) {
Miklos Szeredi22590e42007-10-16 23:27:08 -0700269 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
271 /*
272 * Nobody gets write access to a read-only fs.
273 */
274 if (IS_RDONLY(inode) &&
275 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
276 return -EROFS;
277
278 /*
279 * Nobody gets write access to an immutable file.
280 */
281 if (IS_IMMUTABLE(inode))
282 return -EACCES;
283 }
284
Al Viroacfa4382008-12-04 10:06:33 -0500285 if (inode->i_op->permission)
Nick Pigginb74c79e2011-01-07 17:49:58 +1100286 retval = inode->i_op->permission(inode, mask, 0);
Miklos Szeredif696a362008-07-31 13:41:58 +0200287 else
Nick Pigginb74c79e2011-01-07 17:49:58 +1100288 retval = generic_permission(inode, mask, 0,
289 inode->i_op->check_acl);
Miklos Szeredif696a362008-07-31 13:41:58 +0200290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 if (retval)
292 return retval;
293
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700294 retval = devcgroup_inode_permission(inode, mask);
295 if (retval)
296 return retval;
297
Eric Parisd09ca732010-07-23 11:43:57 -0400298 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800301/**
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800302 * file_permission - check for additional access rights to a given file
303 * @file: file to check access rights for
304 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
305 *
306 * Used to check for read/write/execute permissions on an already opened
307 * file.
308 *
309 * Note:
310 * Do not use this function in new code. All access checks should
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200311 * be done using inode_permission().
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800312 */
313int file_permission(struct file *file, int mask)
314{
Al Virof419a2e2008-07-22 00:07:17 -0400315 return inode_permission(file->f_path.dentry->d_inode, mask);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800316}
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318/*
319 * get_write_access() gets write permission for a file.
320 * put_write_access() releases this write permission.
321 * This is used for regular files.
322 * We cannot support write (and maybe mmap read-write shared) accesses and
323 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
324 * can have the following values:
325 * 0: no writers, no VM_DENYWRITE mappings
326 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
327 * > 0: (i_writecount) users are writing to the file.
328 *
329 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
330 * except for the cases where we don't hold i_writecount yet. Then we need to
331 * use {get,deny}_write_access() - these functions check the sign and refuse
332 * to do the change if sign is wrong. Exclusion between them is provided by
333 * the inode->i_lock spinlock.
334 */
335
336int get_write_access(struct inode * inode)
337{
338 spin_lock(&inode->i_lock);
339 if (atomic_read(&inode->i_writecount) < 0) {
340 spin_unlock(&inode->i_lock);
341 return -ETXTBSY;
342 }
343 atomic_inc(&inode->i_writecount);
344 spin_unlock(&inode->i_lock);
345
346 return 0;
347}
348
349int deny_write_access(struct file * file)
350{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800351 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353 spin_lock(&inode->i_lock);
354 if (atomic_read(&inode->i_writecount) > 0) {
355 spin_unlock(&inode->i_lock);
356 return -ETXTBSY;
357 }
358 atomic_dec(&inode->i_writecount);
359 spin_unlock(&inode->i_lock);
360
361 return 0;
362}
363
Jan Blunck1d957f92008-02-14 19:34:35 -0800364/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800365 * path_get - get a reference to a path
366 * @path: path to get the reference to
367 *
368 * Given a path increment the reference count to the dentry and the vfsmount.
369 */
370void path_get(struct path *path)
371{
372 mntget(path->mnt);
373 dget(path->dentry);
374}
375EXPORT_SYMBOL(path_get);
376
377/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800378 * path_put - put a reference to a path
379 * @path: path to put the reference to
380 *
381 * Given a path decrement the reference count to the dentry and the vfsmount.
382 */
383void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384{
Jan Blunck1d957f92008-02-14 19:34:35 -0800385 dput(path->dentry);
386 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
Jan Blunck1d957f92008-02-14 19:34:35 -0800388EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Trond Myklebust834f2a42005-10-18 14:20:16 -0700390/**
Nick Piggin31e6b012011-01-07 17:49:52 +1100391 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
392 * @nd: nameidata pathwalk data to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800393 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100394 *
395 * Path walking has 2 modes, rcu-walk and ref-walk (see
396 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
397 * to drop out of rcu-walk mode and take normal reference counts on dentries
398 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
399 * refcounts at the last known good point before rcu-walk got stuck, so
400 * ref-walk may continue from there. If this is not successful (eg. a seqcount
401 * has changed), then failure is returned and path walk restarts from the
402 * beginning in ref-walk mode.
403 *
404 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
405 * ref-walk. Must be called from rcu-walk context.
406 */
407static int nameidata_drop_rcu(struct nameidata *nd)
408{
409 struct fs_struct *fs = current->fs;
410 struct dentry *dentry = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500411 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100412
413 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500414 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
415 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100416 spin_lock(&fs->lock);
417 if (nd->root.mnt != fs->root.mnt ||
418 nd->root.dentry != fs->root.dentry)
419 goto err_root;
420 }
421 spin_lock(&dentry->d_lock);
422 if (!__d_rcu_to_refcount(dentry, nd->seq))
423 goto err;
424 BUG_ON(nd->inode != dentry->d_inode);
425 spin_unlock(&dentry->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500426 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100427 path_get(&nd->root);
428 spin_unlock(&fs->lock);
429 }
430 mntget(nd->path.mnt);
431
432 rcu_read_unlock();
433 br_read_unlock(vfsmount_lock);
434 nd->flags &= ~LOOKUP_RCU;
435 return 0;
436err:
437 spin_unlock(&dentry->d_lock);
438err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500439 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100440 spin_unlock(&fs->lock);
441 return -ECHILD;
442}
443
444/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
445static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
446{
447 if (nd->flags & LOOKUP_RCU)
448 return nameidata_drop_rcu(nd);
449 return 0;
450}
451
452/**
453 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
454 * @nd: nameidata pathwalk data to drop
455 * @dentry: dentry to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800456 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100457 *
458 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
459 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
460 * @nd. Must be called from rcu-walk context.
461 */
462static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
463{
464 struct fs_struct *fs = current->fs;
465 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500466 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100467
468 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500469 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
470 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100471 spin_lock(&fs->lock);
472 if (nd->root.mnt != fs->root.mnt ||
473 nd->root.dentry != fs->root.dentry)
474 goto err_root;
475 }
476 spin_lock(&parent->d_lock);
477 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
478 if (!__d_rcu_to_refcount(dentry, nd->seq))
479 goto err;
480 /*
481 * If the sequence check on the child dentry passed, then the child has
482 * not been removed from its parent. This means the parent dentry must
483 * be valid and able to take a reference at this point.
484 */
485 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
486 BUG_ON(!parent->d_count);
487 parent->d_count++;
488 spin_unlock(&dentry->d_lock);
489 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500490 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100491 path_get(&nd->root);
492 spin_unlock(&fs->lock);
493 }
494 mntget(nd->path.mnt);
495
496 rcu_read_unlock();
497 br_read_unlock(vfsmount_lock);
498 nd->flags &= ~LOOKUP_RCU;
499 return 0;
500err:
501 spin_unlock(&dentry->d_lock);
502 spin_unlock(&parent->d_lock);
503err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500504 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100505 spin_unlock(&fs->lock);
506 return -ECHILD;
507}
508
509/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
510static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
511{
Al Viroa7472ba2011-03-04 14:39:30 -0500512 if (nd->flags & LOOKUP_RCU) {
513 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
514 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500515 if (!(nd->flags & LOOKUP_ROOT))
516 nd->root.mnt = NULL;
Al Viroa7472ba2011-03-04 14:39:30 -0500517 rcu_read_unlock();
518 br_read_unlock(vfsmount_lock);
519 return -ECHILD;
520 }
521 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100522 return 0;
523}
524
525/**
526 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
527 * @nd: nameidata pathwalk data to drop
Randy Dunlap39191622011-01-08 19:36:21 -0800528 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100529 *
530 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
531 * nd->path should be the final element of the lookup, so nd->root is discarded.
532 * Must be called from rcu-walk context.
533 */
534static int nameidata_drop_rcu_last(struct nameidata *nd)
535{
536 struct dentry *dentry = nd->path.dentry;
537
538 BUG_ON(!(nd->flags & LOOKUP_RCU));
539 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500540 if (!(nd->flags & LOOKUP_ROOT))
541 nd->root.mnt = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +1100542 spin_lock(&dentry->d_lock);
543 if (!__d_rcu_to_refcount(dentry, nd->seq))
544 goto err_unlock;
545 BUG_ON(nd->inode != dentry->d_inode);
546 spin_unlock(&dentry->d_lock);
547
548 mntget(nd->path.mnt);
549
550 rcu_read_unlock();
551 br_read_unlock(vfsmount_lock);
552
553 return 0;
554
555err_unlock:
556 spin_unlock(&dentry->d_lock);
557 rcu_read_unlock();
558 br_read_unlock(vfsmount_lock);
559 return -ECHILD;
560}
561
Nick Piggin31e6b012011-01-07 17:49:52 +1100562/**
Trond Myklebust834f2a42005-10-18 14:20:16 -0700563 * release_open_intent - free up open intent resources
564 * @nd: pointer to nameidata
565 */
566void release_open_intent(struct nameidata *nd)
567{
Linus Torvalds2dab5972011-02-11 15:53:38 -0800568 struct file *file = nd->intent.open.file;
569
570 if (file && !IS_ERR(file)) {
571 if (file->f_path.dentry == NULL)
572 put_filp(file);
573 else
574 fput(file);
575 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700576}
577
Al Virof60aef72011-02-15 01:35:28 -0500578static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
Nick Piggin34286d62011-01-07 17:49:57 +1100579{
Al Virof60aef72011-02-15 01:35:28 -0500580 return dentry->d_op->d_revalidate(dentry, nd);
Nick Piggin34286d62011-01-07 17:49:57 +1100581}
582
Al Virof5e1c1c2011-02-15 01:32:55 -0500583static struct dentry *
Ian Kentbcdc5e02006-09-27 01:50:44 -0700584do_revalidate(struct dentry *dentry, struct nameidata *nd)
585{
Al Virof5e1c1c2011-02-15 01:32:55 -0500586 int status = d_revalidate(dentry, nd);
Ian Kentbcdc5e02006-09-27 01:50:44 -0700587 if (unlikely(status <= 0)) {
588 /*
589 * The dentry failed validation.
590 * If d_revalidate returned 0 attempt to invalidate
591 * the dentry otherwise d_revalidate is asking us
592 * to return a fail status.
593 */
Nick Piggin34286d62011-01-07 17:49:57 +1100594 if (status < 0) {
Al Virof5e1c1c2011-02-15 01:32:55 -0500595 dput(dentry);
Nick Piggin34286d62011-01-07 17:49:57 +1100596 dentry = ERR_PTR(status);
Al Virof5e1c1c2011-02-15 01:32:55 -0500597 } else if (!d_invalidate(dentry)) {
598 dput(dentry);
599 dentry = NULL;
Ian Kentbcdc5e02006-09-27 01:50:44 -0700600 }
601 }
602 return dentry;
603}
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605/*
Al Viro16c2cd72011-02-22 15:50:10 -0500606 * handle_reval_path - force revalidation of a dentry
Jeff Layton39159de2009-12-07 12:01:50 -0500607 *
608 * In some situations the path walking code will trust dentries without
609 * revalidating them. This causes problems for filesystems that depend on
610 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
611 * (which indicates that it's possible for the dentry to go stale), force
612 * a d_revalidate call before proceeding.
613 *
614 * Returns 0 if the revalidation was successful. If the revalidation fails,
615 * either return the error returned by d_revalidate or -ESTALE if the
616 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
617 * invalidate the dentry. It's up to the caller to handle putting references
618 * to the path if necessary.
619 */
Al Viro16c2cd72011-02-22 15:50:10 -0500620static inline int handle_reval_path(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500621{
Al Viro16c2cd72011-02-22 15:50:10 -0500622 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500623 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500624
Al Viro16c2cd72011-02-22 15:50:10 -0500625 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500626 return 0;
627
Al Viro16c2cd72011-02-22 15:50:10 -0500628 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
629 return 0;
630
631 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
632 return 0;
633
634 /* Note: we do not d_invalidate() */
Nick Piggin34286d62011-01-07 17:49:57 +1100635 status = d_revalidate(dentry, nd);
Jeff Layton39159de2009-12-07 12:01:50 -0500636 if (status > 0)
637 return 0;
638
Al Viro16c2cd72011-02-22 15:50:10 -0500639 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500640 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500641
Jeff Layton39159de2009-12-07 12:01:50 -0500642 return status;
643}
644
645/*
Al Virob75b5082009-12-16 01:01:38 -0500646 * Short-cut version of permission(), for calling on directories
647 * during pathname resolution. Combines parts of permission()
648 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 *
650 * If appropriate, check DAC only. If not appropriate, or
Al Virob75b5082009-12-16 01:01:38 -0500651 * short-cut DAC fails, then call ->permission() to do more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 * complete permission check.
653 */
Nick Pigginb74c79e2011-01-07 17:49:58 +1100654static inline int exec_permission(struct inode *inode, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700656 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700658 if (inode->i_op->permission) {
Nick Pigginb74c79e2011-01-07 17:49:58 +1100659 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
660 } else {
661 ret = acl_permission_check(inode, MAY_EXEC, flags,
662 inode->i_op->check_acl);
Linus Torvaldscb9179e2009-08-28 11:08:31 -0700663 }
Nick Pigginb74c79e2011-01-07 17:49:58 +1100664 if (likely(!ret))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 goto ok;
Nick Pigginb74c79e2011-01-07 17:49:58 +1100666 if (ret == -ECHILD)
Nick Piggin31e6b012011-01-07 17:49:52 +1100667 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Linus Torvaldsf1ac9f62009-08-28 10:53:56 -0700669 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 goto ok;
671
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700672 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673ok:
Nick Pigginb74c79e2011-01-07 17:49:58 +1100674 return security_inode_exec_permission(inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
Al Viro2a737872009-04-07 11:49:53 -0400677static __always_inline void set_root(struct nameidata *nd)
678{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200679 if (!nd->root.mnt)
680 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400681}
682
Al Viro6de88d72009-08-09 01:41:57 +0400683static int link_path_walk(const char *, struct nameidata *);
684
Nick Piggin31e6b012011-01-07 17:49:52 +1100685static __always_inline void set_root_rcu(struct nameidata *nd)
686{
687 if (!nd->root.mnt) {
688 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100689 unsigned seq;
690
691 do {
692 seq = read_seqcount_begin(&fs->seq);
693 nd->root = fs->root;
694 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100695 }
696}
697
Arjan van de Venf1662352006-01-14 13:21:31 -0800698static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
Nick Piggin31e6b012011-01-07 17:49:52 +1100700 int ret;
701
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (IS_ERR(link))
703 goto fail;
704
705 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400706 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800707 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400708 nd->path = nd->root;
709 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500710 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100712 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100713
Nick Piggin31e6b012011-01-07 17:49:52 +1100714 ret = link_path_walk(link, nd);
715 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800717 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 return PTR_ERR(link);
719}
720
Jan Blunck1d957f92008-02-14 19:34:35 -0800721static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700722{
723 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800724 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700725 mntput(path->mnt);
726}
727
Nick Piggin7b9337a2011-01-14 08:42:43 +0000728static inline void path_to_nameidata(const struct path *path,
729 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700730{
Nick Piggin31e6b012011-01-07 17:49:52 +1100731 if (!(nd->flags & LOOKUP_RCU)) {
732 dput(nd->path.dentry);
733 if (nd->path.mnt != path->mnt)
734 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800735 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100736 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800737 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700738}
739
Al Virodef4af32009-12-26 08:37:05 -0500740static __always_inline int
Nick Piggin7b9337a2011-01-14 08:42:43 +0000741__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800742{
743 int error;
Nick Piggin7b9337a2011-01-14 08:42:43 +0000744 struct dentry *dentry = link->dentry;
Ian Kent051d3812006-03-27 01:14:53 -0800745
Al Viro844a3912011-02-15 00:38:26 -0500746 BUG_ON(nd->flags & LOOKUP_RCU);
747
Nick Piggin7b9337a2011-01-14 08:42:43 +0000748 touch_atime(link->mnt, dentry);
Ian Kent051d3812006-03-27 01:14:53 -0800749 nd_set_link(nd, NULL);
750
David Howells87556ef2011-01-14 18:46:46 +0000751 if (link->mnt == nd->path.mnt)
752 mntget(link->mnt);
Nick Piggin31e6b012011-01-07 17:49:52 +1100753
Al Viro36f3b4f2011-02-22 21:24:38 -0500754 error = security_inode_follow_link(link->dentry, nd);
755 if (error) {
756 *p = ERR_PTR(error); /* no ->put_link(), please */
757 path_put(&nd->path);
758 return error;
759 }
760
Al Viro86acdca12009-12-22 23:45:11 -0500761 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500762 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
763 error = PTR_ERR(*p);
764 if (!IS_ERR(*p)) {
Ian Kent051d3812006-03-27 01:14:53 -0800765 char *s = nd_get_link(nd);
766 error = 0;
767 if (s)
768 error = __vfs_follow_link(nd, s);
Al Virobcda7652011-03-13 16:42:14 -0400769 else if (nd->last_type == LAST_BIND) {
Al Viro16c2cd72011-02-22 15:50:10 -0500770 nd->flags |= LOOKUP_JUMPED;
Al Virobcda7652011-03-13 16:42:14 -0400771 if (nd->path.dentry->d_inode->i_op->follow_link) {
772 /* stepped on a _really_ weird one */
773 path_put(&nd->path);
774 error = -ELOOP;
775 }
776 }
Ian Kent051d3812006-03-27 01:14:53 -0800777 }
Ian Kent051d3812006-03-27 01:14:53 -0800778 return error;
779}
780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781/*
782 * This limits recursive symlink follows to 8, while
783 * limiting consecutive symlinks to 40.
784 *
785 * Without that kind of total limit, nasty chains of consecutive
786 * symlinks can cause almost arbitrarily long lookups.
787 */
Al Viroce57dfc2011-03-13 19:58:58 -0400788static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
Al Virodef4af32009-12-26 08:37:05 -0500790 void *cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 int err = -ELOOP;
Al Viro844a3912011-02-15 00:38:26 -0500792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 if (current->link_count >= MAX_NESTED_LINKS)
794 goto loop;
795 if (current->total_link_count >= 40)
796 goto loop;
797 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
798 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 current->link_count++;
800 current->total_link_count++;
801 nd->depth++;
Al Virodef4af32009-12-26 08:37:05 -0500802 err = __do_follow_link(path, nd, &cookie);
803 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
804 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
Al Viro258fa992009-08-09 01:32:02 +0400805 path_put(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 current->link_count--;
807 nd->depth--;
808 return err;
809loop:
Jan Blunck1d957f92008-02-14 19:34:35 -0800810 path_put_conditional(path, nd);
811 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 return err;
813}
814
Nick Piggin31e6b012011-01-07 17:49:52 +1100815static int follow_up_rcu(struct path *path)
816{
817 struct vfsmount *parent;
818 struct dentry *mountpoint;
819
820 parent = path->mnt->mnt_parent;
821 if (parent == path->mnt)
822 return 0;
823 mountpoint = path->mnt->mnt_mountpoint;
824 path->dentry = mountpoint;
825 path->mnt = parent;
826 return 1;
827}
828
Al Virobab77eb2009-04-18 03:26:48 -0400829int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
831 struct vfsmount *parent;
832 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000833
834 br_read_lock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400835 parent = path->mnt->mnt_parent;
836 if (parent == path->mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +1000837 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 return 0;
839 }
840 mntget(parent);
Al Virobab77eb2009-04-18 03:26:48 -0400841 mountpoint = dget(path->mnt->mnt_mountpoint);
Nick Piggin99b7db72010-08-18 04:37:39 +1000842 br_read_unlock(vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400843 dput(path->dentry);
844 path->dentry = mountpoint;
845 mntput(path->mnt);
846 path->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 return 1;
848}
849
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100850/*
David Howells9875cf82011-01-14 18:45:21 +0000851 * Perform an automount
852 * - return -EISDIR to tell follow_managed() to stop and return the path we
853 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 */
David Howells9875cf82011-01-14 18:45:21 +0000855static int follow_automount(struct path *path, unsigned flags,
856 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100857{
David Howells9875cf82011-01-14 18:45:21 +0000858 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000859 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100860
David Howells9875cf82011-01-14 18:45:21 +0000861 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
862 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700863
David Howells6f45b652011-01-14 18:45:31 +0000864 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
865 * and this is the terminal part of the path.
866 */
867 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
868 return -EISDIR; /* we actually want to stop here */
869
David Howells9875cf82011-01-14 18:45:21 +0000870 /* We want to mount if someone is trying to open/create a file of any
871 * type under the mountpoint, wants to traverse through the mountpoint
872 * or wants to open the mounted directory.
873 *
874 * We don't want to mount if someone's just doing a stat and they've
875 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
876 * appended a '/' to the name.
877 */
878 if (!(flags & LOOKUP_FOLLOW) &&
879 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
880 LOOKUP_OPEN | LOOKUP_CREATE)))
881 return -EISDIR;
882
883 current->total_link_count++;
884 if (current->total_link_count >= 40)
885 return -ELOOP;
886
887 mnt = path->dentry->d_op->d_automount(path);
888 if (IS_ERR(mnt)) {
889 /*
890 * The filesystem is allowed to return -EISDIR here to indicate
891 * it doesn't want to automount. For instance, autofs would do
892 * this so that its userspace daemon can mount on this dentry.
893 *
894 * However, we can only permit this if it's a terminal point in
895 * the path being looked up; if it wasn't then the remainder of
896 * the path is inaccessible and we should say so.
897 */
898 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
899 return -EREMOTE;
900 return PTR_ERR(mnt);
901 }
David Howellsea5b7782011-01-14 19:10:03 +0000902
David Howells9875cf82011-01-14 18:45:21 +0000903 if (!mnt) /* mount collision */
904 return 0;
905
Al Viro19a167a2011-01-17 01:35:23 -0500906 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000907
David Howellsea5b7782011-01-14 19:10:03 +0000908 switch (err) {
909 case -EBUSY:
910 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500911 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000912 case 0:
David Howellsea5b7782011-01-14 19:10:03 +0000913 dput(path->dentry);
914 if (*need_mntput)
915 mntput(path->mnt);
916 path->mnt = mnt;
917 path->dentry = dget(mnt->mnt_root);
918 *need_mntput = true;
919 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500920 default:
921 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000922 }
Al Viro19a167a2011-01-17 01:35:23 -0500923
David Howells9875cf82011-01-14 18:45:21 +0000924}
925
926/*
927 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000928 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000929 * - Flagged as mountpoint
930 * - Flagged as automount point
931 *
932 * This may only be called in refwalk mode.
933 *
934 * Serialization is taken care of in namespace.c
935 */
936static int follow_managed(struct path *path, unsigned flags)
937{
938 unsigned managed;
939 bool need_mntput = false;
940 int ret;
941
942 /* Given that we're not holding a lock here, we retain the value in a
943 * local variable for each dentry as we look at it so that we don't see
944 * the components of that value change under us */
945 while (managed = ACCESS_ONCE(path->dentry->d_flags),
946 managed &= DCACHE_MANAGED_DENTRY,
947 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000948 /* Allow the filesystem to manage the transit without i_mutex
949 * being held. */
950 if (managed & DCACHE_MANAGE_TRANSIT) {
951 BUG_ON(!path->dentry->d_op);
952 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +0000953 ret = path->dentry->d_op->d_manage(path->dentry,
954 false, false);
David Howellscc53ce52011-01-14 18:45:26 +0000955 if (ret < 0)
956 return ret == -EISDIR ? 0 : ret;
957 }
958
David Howells9875cf82011-01-14 18:45:21 +0000959 /* Transit to a mounted filesystem. */
960 if (managed & DCACHE_MOUNTED) {
961 struct vfsmount *mounted = lookup_mnt(path);
962 if (mounted) {
963 dput(path->dentry);
964 if (need_mntput)
965 mntput(path->mnt);
966 path->mnt = mounted;
967 path->dentry = dget(mounted->mnt_root);
968 need_mntput = true;
969 continue;
970 }
971
972 /* Something is mounted on this dentry in another
973 * namespace and/or whatever was mounted there in this
974 * namespace got unmounted before we managed to get the
975 * vfsmount_lock */
976 }
977
978 /* Handle an automount point */
979 if (managed & DCACHE_NEED_AUTOMOUNT) {
980 ret = follow_automount(path, flags, &need_mntput);
981 if (ret < 0)
982 return ret == -EISDIR ? 0 : ret;
983 continue;
984 }
985
986 /* We didn't change the current path point */
987 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 }
David Howells9875cf82011-01-14 18:45:21 +0000989 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990}
991
David Howellscc53ce52011-01-14 18:45:26 +0000992int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993{
994 struct vfsmount *mounted;
995
Al Viro1c755af2009-04-18 14:06:57 -0400996 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400998 dput(path->dentry);
999 mntput(path->mnt);
1000 path->mnt = mounted;
1001 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 return 1;
1003 }
1004 return 0;
1005}
1006
David Howells9875cf82011-01-14 18:45:21 +00001007/*
1008 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
David Howellscc53ce52011-01-14 18:45:26 +00001009 * meet a managed dentry and we're not walking to "..". True is returned to
David Howells9875cf82011-01-14 18:45:21 +00001010 * continue, false to abort.
1011 */
1012static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1013 struct inode **inode, bool reverse_transit)
1014{
1015 while (d_mountpoint(path->dentry)) {
1016 struct vfsmount *mounted;
David Howellsab909112011-01-14 18:46:51 +00001017 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1018 !reverse_transit &&
1019 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1020 return false;
David Howells9875cf82011-01-14 18:45:21 +00001021 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1022 if (!mounted)
1023 break;
1024 path->mnt = mounted;
1025 path->dentry = mounted->mnt_root;
1026 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1027 *inode = path->dentry->d_inode;
1028 }
1029
1030 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1031 return reverse_transit;
1032 return true;
1033}
1034
Nick Piggin31e6b012011-01-07 17:49:52 +11001035static int follow_dotdot_rcu(struct nameidata *nd)
1036{
1037 struct inode *inode = nd->inode;
1038
1039 set_root_rcu(nd);
1040
David Howells9875cf82011-01-14 18:45:21 +00001041 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001042 if (nd->path.dentry == nd->root.dentry &&
1043 nd->path.mnt == nd->root.mnt) {
1044 break;
1045 }
1046 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1047 struct dentry *old = nd->path.dentry;
1048 struct dentry *parent = old->d_parent;
1049 unsigned seq;
1050
1051 seq = read_seqcount_begin(&parent->d_seq);
1052 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -05001053 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +11001054 inode = parent->d_inode;
1055 nd->path.dentry = parent;
1056 nd->seq = seq;
1057 break;
1058 }
1059 if (!follow_up_rcu(&nd->path))
1060 break;
1061 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1062 inode = nd->path.dentry->d_inode;
1063 }
David Howells9875cf82011-01-14 18:45:21 +00001064 __follow_mount_rcu(nd, &nd->path, &inode, true);
Nick Piggin31e6b012011-01-07 17:49:52 +11001065 nd->inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001066 return 0;
Al Viroef7562d2011-03-04 14:35:59 -05001067
1068failed:
1069 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001070 if (!(nd->flags & LOOKUP_ROOT))
1071 nd->root.mnt = NULL;
Al Viroef7562d2011-03-04 14:35:59 -05001072 rcu_read_unlock();
1073 br_read_unlock(vfsmount_lock);
1074 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001075}
1076
David Howells9875cf82011-01-14 18:45:21 +00001077/*
David Howellscc53ce52011-01-14 18:45:26 +00001078 * Follow down to the covering mount currently visible to userspace. At each
1079 * point, the filesystem owning that dentry may be queried as to whether the
1080 * caller is permitted to proceed or not.
1081 *
1082 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1083 * being true).
1084 */
1085int follow_down(struct path *path, bool mounting_here)
1086{
1087 unsigned managed;
1088 int ret;
1089
1090 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1091 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1092 /* Allow the filesystem to manage the transit without i_mutex
1093 * being held.
1094 *
1095 * We indicate to the filesystem if someone is trying to mount
1096 * something here. This gives autofs the chance to deny anyone
1097 * other than its daemon the right to mount on its
1098 * superstructure.
1099 *
1100 * The filesystem may sleep at this point.
1101 */
1102 if (managed & DCACHE_MANAGE_TRANSIT) {
1103 BUG_ON(!path->dentry->d_op);
1104 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001105 ret = path->dentry->d_op->d_manage(
1106 path->dentry, mounting_here, false);
David Howellscc53ce52011-01-14 18:45:26 +00001107 if (ret < 0)
1108 return ret == -EISDIR ? 0 : ret;
1109 }
1110
1111 /* Transit to a mounted filesystem. */
1112 if (managed & DCACHE_MOUNTED) {
1113 struct vfsmount *mounted = lookup_mnt(path);
1114 if (!mounted)
1115 break;
1116 dput(path->dentry);
1117 mntput(path->mnt);
1118 path->mnt = mounted;
1119 path->dentry = dget(mounted->mnt_root);
1120 continue;
1121 }
1122
1123 /* Don't handle automount points here */
1124 break;
1125 }
1126 return 0;
1127}
1128
1129/*
David Howells9875cf82011-01-14 18:45:21 +00001130 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1131 */
1132static void follow_mount(struct path *path)
1133{
1134 while (d_mountpoint(path->dentry)) {
1135 struct vfsmount *mounted = lookup_mnt(path);
1136 if (!mounted)
1137 break;
1138 dput(path->dentry);
1139 mntput(path->mnt);
1140 path->mnt = mounted;
1141 path->dentry = dget(mounted->mnt_root);
1142 }
1143}
1144
Nick Piggin31e6b012011-01-07 17:49:52 +11001145static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146{
Al Viro2a737872009-04-07 11:49:53 -04001147 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001150 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
Al Viro2a737872009-04-07 11:49:53 -04001152 if (nd->path.dentry == nd->root.dentry &&
1153 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 break;
1155 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001156 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001157 /* rare case of legitimate dget_parent()... */
1158 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 dput(old);
1160 break;
1161 }
Al Viro3088dd72010-01-30 15:47:29 -05001162 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 }
Al Viro79ed0222009-04-18 13:59:41 -04001165 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001166 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167}
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169/*
Nick Pigginbaa03892010-08-18 04:37:31 +10001170 * Allocate a dentry with name and parent, and perform a parent
1171 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1172 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1173 * have verified that no child exists while under i_mutex.
1174 */
1175static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1176 struct qstr *name, struct nameidata *nd)
1177{
1178 struct inode *inode = parent->d_inode;
1179 struct dentry *dentry;
1180 struct dentry *old;
1181
1182 /* Don't create child dentry for a dead directory. */
1183 if (unlikely(IS_DEADDIR(inode)))
1184 return ERR_PTR(-ENOENT);
1185
1186 dentry = d_alloc(parent, name);
1187 if (unlikely(!dentry))
1188 return ERR_PTR(-ENOMEM);
1189
1190 old = inode->i_op->lookup(inode, dentry, nd);
1191 if (unlikely(old)) {
1192 dput(dentry);
1193 dentry = old;
1194 }
1195 return dentry;
1196}
1197
1198/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 * It's more convoluted than I'd like it to be, but... it's still fairly
1200 * small and for now I'd prefer to have fast path as straight as possible.
1201 * It _is_ time-critical.
1202 */
1203static int do_lookup(struct nameidata *nd, struct qstr *name,
Nick Piggin31e6b012011-01-07 17:49:52 +11001204 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
Jan Blunck4ac91372008-02-14 19:34:32 -08001206 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001207 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001208 int need_reval = 1;
1209 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001210 int err;
1211
Al Viro3cac2602009-08-13 18:27:43 +04001212 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001213 * Rename seqlock is not required here because in the off chance
1214 * of a false negative due to a concurrent rename, we're going to
1215 * do the non-racy lookup, below.
1216 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001217 if (nd->flags & LOOKUP_RCU) {
1218 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001219 *inode = nd->inode;
1220 dentry = __d_lookup_rcu(parent, name, &seq, inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001221 if (!dentry)
1222 goto unlazy;
1223
Nick Piggin31e6b012011-01-07 17:49:52 +11001224 /* Memory barrier in read_seqcount_begin of child is enough */
1225 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1226 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001227 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001228
Al Viro24643082011-02-15 01:26:22 -05001229 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro5a18fff2011-03-11 04:44:53 -05001230 status = d_revalidate(dentry, nd);
1231 if (unlikely(status <= 0)) {
1232 if (status != -ECHILD)
1233 need_reval = 0;
1234 goto unlazy;
1235 }
Al Viro24643082011-02-15 01:26:22 -05001236 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001237 path->mnt = mnt;
1238 path->dentry = dentry;
David Howells9875cf82011-01-14 18:45:21 +00001239 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1240 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001241unlazy:
1242 if (dentry) {
1243 if (nameidata_dentry_drop_rcu(nd, dentry))
1244 return -ECHILD;
1245 } else {
1246 if (nameidata_drop_rcu(nd))
1247 return -ECHILD;
1248 }
1249 } else {
1250 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001251 }
Al Viro5a18fff2011-03-11 04:44:53 -05001252
1253retry:
1254 if (unlikely(!dentry)) {
1255 struct inode *dir = parent->d_inode;
1256 BUG_ON(nd->inode != dir);
1257
1258 mutex_lock(&dir->i_mutex);
1259 dentry = d_lookup(parent, name);
1260 if (likely(!dentry)) {
1261 dentry = d_alloc_and_lookup(parent, name, nd);
1262 if (IS_ERR(dentry)) {
1263 mutex_unlock(&dir->i_mutex);
1264 return PTR_ERR(dentry);
1265 }
1266 /* known good */
1267 need_reval = 0;
1268 status = 1;
1269 }
1270 mutex_unlock(&dir->i_mutex);
Al Viro24643082011-02-15 01:26:22 -05001271 }
Al Viro5a18fff2011-03-11 04:44:53 -05001272 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1273 status = d_revalidate(dentry, nd);
1274 if (unlikely(status <= 0)) {
1275 if (status < 0) {
1276 dput(dentry);
1277 return status;
1278 }
1279 if (!d_invalidate(dentry)) {
1280 dput(dentry);
1281 dentry = NULL;
1282 need_reval = 1;
1283 goto retry;
1284 }
1285 }
1286
David Howells9875cf82011-01-14 18:45:21 +00001287 path->mnt = mnt;
1288 path->dentry = dentry;
1289 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001290 if (unlikely(err < 0)) {
1291 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001292 return err;
Ian Kent89312212011-01-18 12:06:10 +08001293 }
David Howells9875cf82011-01-14 18:45:21 +00001294 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296}
1297
Al Viro52094c82011-02-21 21:34:47 -05001298static inline int may_lookup(struct nameidata *nd)
1299{
1300 if (nd->flags & LOOKUP_RCU) {
1301 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1302 if (err != -ECHILD)
1303 return err;
1304 if (nameidata_drop_rcu(nd))
1305 return -ECHILD;
1306 }
1307 return exec_permission(nd->inode, 0);
1308}
1309
Al Viro9856fa12011-03-04 14:22:06 -05001310static inline int handle_dots(struct nameidata *nd, int type)
1311{
1312 if (type == LAST_DOTDOT) {
1313 if (nd->flags & LOOKUP_RCU) {
1314 if (follow_dotdot_rcu(nd))
1315 return -ECHILD;
1316 } else
1317 follow_dotdot(nd);
1318 }
1319 return 0;
1320}
1321
Al Viro951361f2011-03-04 14:44:37 -05001322static void terminate_walk(struct nameidata *nd)
1323{
1324 if (!(nd->flags & LOOKUP_RCU)) {
1325 path_put(&nd->path);
1326 } else {
1327 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001328 if (!(nd->flags & LOOKUP_ROOT))
1329 nd->root.mnt = NULL;
Al Viro951361f2011-03-04 14:44:37 -05001330 rcu_read_unlock();
1331 br_read_unlock(vfsmount_lock);
1332 }
1333}
1334
Al Viroce57dfc2011-03-13 19:58:58 -04001335static inline int walk_component(struct nameidata *nd, struct path *path,
1336 struct qstr *name, int type, int follow)
1337{
1338 struct inode *inode;
1339 int err;
1340 /*
1341 * "." and ".." are special - ".." especially so because it has
1342 * to be able to know about the current root directory and
1343 * parent relationships.
1344 */
1345 if (unlikely(type != LAST_NORM))
1346 return handle_dots(nd, type);
1347 err = do_lookup(nd, name, path, &inode);
1348 if (unlikely(err)) {
1349 terminate_walk(nd);
1350 return err;
1351 }
1352 if (!inode) {
1353 path_to_nameidata(path, nd);
1354 terminate_walk(nd);
1355 return -ENOENT;
1356 }
1357 if (unlikely(inode->i_op->follow_link) && follow) {
1358 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1359 return -ECHILD;
1360 BUG_ON(inode != path->dentry->d_inode);
1361 return 1;
1362 }
1363 path_to_nameidata(path, nd);
1364 nd->inode = inode;
1365 return 0;
1366}
1367
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368/*
1369 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001370 * This is the basic name resolution function, turning a pathname into
1371 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001373 * Returns 0 and nd will have valid dentry and mnt on success.
1374 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 */
Al Viro6de88d72009-08-09 01:41:57 +04001376static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
1378 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 int err;
1380 unsigned int lookup_flags = nd->flags;
1381
1382 while (*name=='/')
1383 name++;
1384 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001385 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -08001388 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
1390 /* At this point we know we have a real path component. */
1391 for(;;) {
1392 unsigned long hash;
1393 struct qstr this;
1394 unsigned int c;
Al Virofe479a52011-02-22 15:10:03 -05001395 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
Trond Myklebustcdce5d62005-10-18 14:20:18 -07001397 nd->flags |= LOOKUP_CONTINUE;
Al Viro52094c82011-02-21 21:34:47 -05001398
1399 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 if (err)
1401 break;
1402
1403 this.name = name;
1404 c = *(const unsigned char *)name;
1405
1406 hash = init_name_hash();
1407 do {
1408 name++;
1409 hash = partial_name_hash(c, hash);
1410 c = *(const unsigned char *)name;
1411 } while (c && (c != '/'));
1412 this.len = name - (const char *) this.name;
1413 this.hash = end_name_hash(hash);
1414
Al Virofe479a52011-02-22 15:10:03 -05001415 type = LAST_NORM;
1416 if (this.name[0] == '.') switch (this.len) {
1417 case 2:
Al Viro16c2cd72011-02-22 15:50:10 -05001418 if (this.name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001419 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001420 nd->flags |= LOOKUP_JUMPED;
1421 }
Al Virofe479a52011-02-22 15:10:03 -05001422 break;
1423 case 1:
1424 type = LAST_DOT;
1425 }
Al Viro5a202bc2011-03-08 14:17:44 -05001426 if (likely(type == LAST_NORM)) {
1427 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001428 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001429 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1430 err = parent->d_op->d_hash(parent, nd->inode,
1431 &this);
1432 if (err < 0)
1433 break;
1434 }
1435 }
Al Virofe479a52011-02-22 15:10:03 -05001436
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 /* remove trailing slashes? */
1438 if (!c)
1439 goto last_component;
1440 while (*++name == '/');
1441 if (!*name)
1442 goto last_with_slashes;
1443
Al Viroce57dfc2011-03-13 19:58:58 -04001444 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1445 if (err < 0)
1446 return err;
Al Virofe479a52011-02-22 15:10:03 -05001447
Al Viroce57dfc2011-03-13 19:58:58 -04001448 if (err) {
1449 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001451 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001452 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 err = -ENOTDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11001455 if (!nd->inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 break;
1457 continue;
1458 /* here ends the main loop */
1459
1460last_with_slashes:
1461 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1462last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -08001463 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1464 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Al Viroce57dfc2011-03-13 19:58:58 -04001465 if (lookup_flags & LOOKUP_PARENT) {
1466 nd->last = this;
1467 nd->last_type = type;
1468 return 0;
1469 }
1470 err = walk_component(nd, &next, &this, type,
1471 lookup_flags & LOOKUP_FOLLOW);
1472 if (err < 0)
1473 return err;
1474 if (err) {
1475 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001477 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001478 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001479 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 if (lookup_flags & LOOKUP_DIRECTORY) {
1481 err = -ENOTDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11001482 if (!nd->inode->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 break;
1484 }
Al Viro086e1832011-02-22 20:56:27 -05001485 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 }
Al Viro951361f2011-03-04 14:44:37 -05001487 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 return err;
1489}
1490
Al Viro70e9b352011-03-05 21:12:22 -05001491static int path_init(int dfd, const char *name, unsigned int flags,
1492 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001494 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001495 int fput_needed;
1496 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001499 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001501 if (flags & LOOKUP_ROOT) {
1502 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001503 if (*name) {
1504 if (!inode->i_op->lookup)
1505 return -ENOTDIR;
1506 retval = inode_permission(inode, MAY_EXEC);
1507 if (retval)
1508 return retval;
1509 }
Al Viro5b6ca022011-03-09 23:04:47 -05001510 nd->path = nd->root;
1511 nd->inode = inode;
1512 if (flags & LOOKUP_RCU) {
1513 br_read_lock(vfsmount_lock);
1514 rcu_read_lock();
1515 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1516 } else {
1517 path_get(&nd->path);
1518 }
1519 return 0;
1520 }
1521
Al Viro2a737872009-04-07 11:49:53 -04001522 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001525 if (flags & LOOKUP_RCU) {
1526 br_read_lock(vfsmount_lock);
1527 rcu_read_lock();
1528 set_root_rcu(nd);
1529 } else {
1530 set_root(nd);
1531 path_get(&nd->root);
1532 }
Al Viro2a737872009-04-07 11:49:53 -04001533 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001534 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001535 if (flags & LOOKUP_RCU) {
1536 struct fs_struct *fs = current->fs;
1537 unsigned seq;
1538
1539 br_read_lock(vfsmount_lock);
1540 rcu_read_lock();
1541
1542 do {
1543 seq = read_seqcount_begin(&fs->seq);
1544 nd->path = fs->pwd;
1545 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1546 } while (read_seqcount_retry(&fs->seq, seq));
1547 } else {
1548 get_fs_pwd(current->fs, &nd->path);
1549 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001550 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001551 struct dentry *dentry;
1552
Al Viro1abf0c72011-03-13 03:51:11 -04001553 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001554 retval = -EBADF;
1555 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001556 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001557
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001558 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001559
Al Virof52e0c12011-03-14 18:56:51 -04001560 if (*name) {
1561 retval = -ENOTDIR;
1562 if (!S_ISDIR(dentry->d_inode->i_mode))
1563 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001564
Al Virof52e0c12011-03-14 18:56:51 -04001565 retval = file_permission(file, MAY_EXEC);
1566 if (retval)
1567 goto fput_fail;
1568 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001569
Jan Blunck5dd784d02008-02-14 19:34:38 -08001570 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001571 if (flags & LOOKUP_RCU) {
1572 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001573 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001574 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1575 br_read_lock(vfsmount_lock);
1576 rcu_read_lock();
1577 } else {
1578 path_get(&file->f_path);
1579 fput_light(file, fput_needed);
1580 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 }
Al Viroe41f7d42011-02-22 14:02:58 -05001582
Nick Piggin31e6b012011-01-07 17:49:52 +11001583 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001584 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001585
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001586fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001587 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001588out_fail:
1589 return retval;
1590}
1591
1592/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001593static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001594 unsigned int flags, struct nameidata *nd)
1595{
Al Viro70e9b352011-03-05 21:12:22 -05001596 struct file *base = NULL;
Nick Piggin31e6b012011-01-07 17:49:52 +11001597 int retval;
1598
1599 /*
1600 * Path walking is largely split up into 2 different synchronisation
1601 * schemes, rcu-walk and ref-walk (explained in
1602 * Documentation/filesystems/path-lookup.txt). These share much of the
1603 * path walk code, but some things particularly setup, cleanup, and
1604 * following mounts are sufficiently divergent that functions are
1605 * duplicated. Typically there is a function foo(), and its RCU
1606 * analogue, foo_rcu().
1607 *
1608 * -ECHILD is the error number of choice (just to avoid clashes) that
1609 * is returned if some aspect of an rcu-walk fails. Such an error must
1610 * be handled by restarting a traditional ref-walk (which will always
1611 * be able to complete).
1612 */
Al Viro70e9b352011-03-05 21:12:22 -05001613 retval = path_init(dfd, name, flags, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001614
Nick Piggin31e6b012011-01-07 17:49:52 +11001615 if (unlikely(retval))
1616 return retval;
Al Viroee0827c2011-02-21 23:38:09 -05001617
1618 current->total_link_count = 0;
1619 retval = link_path_walk(name, nd);
1620
1621 if (nd->flags & LOOKUP_RCU) {
Al Viro4455ca62011-03-04 14:28:10 -05001622 /* went all way through without dropping RCU */
1623 BUG_ON(retval);
1624 if (nameidata_drop_rcu_last(nd))
1625 retval = -ECHILD;
Al Viroee0827c2011-02-21 23:38:09 -05001626 }
1627
Al Viro16c2cd72011-02-22 15:50:10 -05001628 if (!retval)
1629 retval = handle_reval_path(nd);
1630
Al Viro70e9b352011-03-05 21:12:22 -05001631 if (base)
1632 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001633
Al Viro5b6ca022011-03-09 23:04:47 -05001634 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001635 path_put(&nd->root);
1636 nd->root.mnt = NULL;
1637 }
Al Viroee0827c2011-02-21 23:38:09 -05001638 return retval;
1639}
Nick Piggin31e6b012011-01-07 17:49:52 +11001640
Al Viroee0827c2011-02-21 23:38:09 -05001641static int do_path_lookup(int dfd, const char *name,
1642 unsigned int flags, struct nameidata *nd)
1643{
1644 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1645 if (unlikely(retval == -ECHILD))
1646 retval = path_lookupat(dfd, name, flags, nd);
1647 if (unlikely(retval == -ESTALE))
1648 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001649
1650 if (likely(!retval)) {
1651 if (unlikely(!audit_dummy_context())) {
1652 if (nd->path.dentry && nd->inode)
1653 audit_inode(name, nd->path.dentry);
1654 }
1655 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001656 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657}
1658
Al Viroc9c6cac2011-02-16 15:15:47 -05001659int kern_path_parent(const char *name, struct nameidata *nd)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001660{
Al Viroc9c6cac2011-02-16 15:15:47 -05001661 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001662}
1663
Al Virod1811462008-08-02 00:49:18 -04001664int kern_path(const char *name, unsigned int flags, struct path *path)
1665{
1666 struct nameidata nd;
1667 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1668 if (!res)
1669 *path = nd.path;
1670 return res;
1671}
1672
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001673/**
1674 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1675 * @dentry: pointer to dentry of the base directory
1676 * @mnt: pointer to vfs mount of the base directory
1677 * @name: pointer to file name
1678 * @flags: lookup flags
1679 * @nd: pointer to nameidata
1680 */
1681int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1682 const char *name, unsigned int flags,
1683 struct nameidata *nd)
1684{
Al Viro5b6ca022011-03-09 23:04:47 -05001685 nd->root.dentry = dentry;
1686 nd->root.mnt = mnt;
1687 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1688 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001689}
1690
Christoph Hellwigeead1912007-10-16 23:25:38 -07001691static struct dentry *__lookup_hash(struct qstr *name,
1692 struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693{
Christoph Hellwig81fca442010-10-06 10:47:47 +02001694 struct inode *inode = base->d_inode;
James Morris057f6c02007-04-26 00:12:05 -07001695 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 int err;
1697
Nick Pigginb74c79e2011-01-07 17:49:58 +11001698 err = exec_permission(inode, 0);
Christoph Hellwig81fca442010-10-06 10:47:47 +02001699 if (err)
1700 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
1702 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001703 * Don't bother with __d_lookup: callers are for creat as
1704 * well as unlink, so a lot of the time it would cost
1705 * a double lookup.
Al Viro6e6b1bd2009-08-13 23:38:37 +04001706 */
Nick Pigginb04f7842010-08-18 04:37:34 +10001707 dentry = d_lookup(base, name);
Al Viro6e6b1bd2009-08-13 23:38:37 +04001708
Nick Pigginfb045ad2011-01-07 17:49:55 +11001709 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
Al Viro6e6b1bd2009-08-13 23:38:37 +04001710 dentry = do_revalidate(dentry, nd);
1711
Nick Pigginbaa03892010-08-18 04:37:31 +10001712 if (!dentry)
1713 dentry = d_alloc_and_lookup(base, name, nd);
Al Viro5a202bc2011-03-08 14:17:44 -05001714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 return dentry;
1716}
1717
James Morris057f6c02007-04-26 00:12:05 -07001718/*
1719 * Restricted form of lookup. Doesn't follow links, single-component only,
1720 * needs parent already locked. Doesn't follow mounts.
1721 * SMP-safe.
1722 */
Adrian Bunka244e162006-03-31 02:32:11 -08001723static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724{
Jan Blunck4ac91372008-02-14 19:34:32 -08001725 return __lookup_hash(&nd->last, nd->path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726}
1727
Christoph Hellwigeead1912007-10-16 23:25:38 -07001728/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001729 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001730 * @name: pathname component to lookup
1731 * @base: base directory to lookup from
1732 * @len: maximum length @len should be interpreted to
1733 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001734 * Note that this routine is purely a helper for filesystem usage and should
1735 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001736 * nameidata argument is passed to the filesystem methods and a filesystem
1737 * using this helper needs to be prepared for that.
1738 */
James Morris057f6c02007-04-26 00:12:05 -07001739struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1740{
James Morris057f6c02007-04-26 00:12:05 -07001741 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001742 unsigned long hash;
1743 unsigned int c;
James Morris057f6c02007-04-26 00:12:05 -07001744
David Woodhouse2f9092e2009-04-20 23:18:37 +01001745 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1746
Al Viro6a96ba52011-03-07 23:49:20 -05001747 this.name = name;
1748 this.len = len;
1749 if (!len)
1750 return ERR_PTR(-EACCES);
1751
1752 hash = init_name_hash();
1753 while (len--) {
1754 c = *(const unsigned char *)name++;
1755 if (c == '/' || c == '\0')
1756 return ERR_PTR(-EACCES);
1757 hash = partial_name_hash(c, hash);
1758 }
1759 this.hash = end_name_hash(hash);
Al Viro5a202bc2011-03-08 14:17:44 -05001760 /*
1761 * See if the low-level filesystem might want
1762 * to use its own hash..
1763 */
1764 if (base->d_flags & DCACHE_OP_HASH) {
1765 int err = base->d_op->d_hash(base, base->d_inode, &this);
1766 if (err < 0)
1767 return ERR_PTR(err);
1768 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001769
Christoph Hellwig49705b72005-11-08 21:35:06 -08001770 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001771}
1772
Al Viro2d8f3032008-07-22 09:59:21 -04001773int user_path_at(int dfd, const char __user *name, unsigned flags,
1774 struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775{
Al Viro2d8f3032008-07-22 09:59:21 -04001776 struct nameidata nd;
Al Virof52e0c12011-03-14 18:56:51 -04001777 char *tmp = getname_flags(name, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001780
1781 BUG_ON(flags & LOOKUP_PARENT);
1782
1783 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001785 if (!err)
1786 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 }
1788 return err;
1789}
1790
Al Viro2ad94ae2008-07-21 09:32:51 -04001791static int user_path_parent(int dfd, const char __user *path,
1792 struct nameidata *nd, char **name)
1793{
1794 char *s = getname(path);
1795 int error;
1796
1797 if (IS_ERR(s))
1798 return PTR_ERR(s);
1799
1800 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1801 if (error)
1802 putname(s);
1803 else
1804 *name = s;
1805
1806 return error;
1807}
1808
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809/*
1810 * It's inline, so penalty for filesystems that don't use sticky bit is
1811 * minimal.
1812 */
1813static inline int check_sticky(struct inode *dir, struct inode *inode)
1814{
David Howellsda9592e2008-11-14 10:39:05 +11001815 uid_t fsuid = current_fsuid();
1816
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 if (!(dir->i_mode & S_ISVTX))
1818 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001819 if (inode->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 return 0;
David Howellsda9592e2008-11-14 10:39:05 +11001821 if (dir->i_uid == fsuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 return 0;
1823 return !capable(CAP_FOWNER);
1824}
1825
1826/*
1827 * Check whether we can remove a link victim from directory dir, check
1828 * whether the type of victim is right.
1829 * 1. We can't do it if dir is read-only (done in permission())
1830 * 2. We should have write and exec permissions on dir
1831 * 3. We can't remove anything from append-only dir
1832 * 4. We can't do anything with immutable dir (done in permission())
1833 * 5. If the sticky bit on dir is set we should either
1834 * a. be owner of dir, or
1835 * b. be owner of victim, or
1836 * c. have CAP_FOWNER capability
1837 * 6. If the victim is append-only or immutable we can't do antyhing with
1838 * links pointing to it.
1839 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1840 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1841 * 9. We can't remove a root or mountpoint.
1842 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1843 * nfs_async_unlink().
1844 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001845static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846{
1847 int error;
1848
1849 if (!victim->d_inode)
1850 return -ENOENT;
1851
1852 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05001853 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Al Virof419a2e2008-07-22 00:07:17 -04001855 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 if (error)
1857 return error;
1858 if (IS_APPEND(dir))
1859 return -EPERM;
1860 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08001861 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 return -EPERM;
1863 if (isdir) {
1864 if (!S_ISDIR(victim->d_inode->i_mode))
1865 return -ENOTDIR;
1866 if (IS_ROOT(victim))
1867 return -EBUSY;
1868 } else if (S_ISDIR(victim->d_inode->i_mode))
1869 return -EISDIR;
1870 if (IS_DEADDIR(dir))
1871 return -ENOENT;
1872 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1873 return -EBUSY;
1874 return 0;
1875}
1876
1877/* Check whether we can create an object with dentry child in directory
1878 * dir.
1879 * 1. We can't do it if child already exists (open has special treatment for
1880 * this case, but since we are inlined it's OK)
1881 * 2. We can't do it if dir is read-only (done in permission())
1882 * 3. We should have write and exec permissions on dir
1883 * 4. We can't do it if dir is immutable (done in permission())
1884 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02001885static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886{
1887 if (child->d_inode)
1888 return -EEXIST;
1889 if (IS_DEADDIR(dir))
1890 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04001891 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892}
1893
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894/*
1895 * p1 and p2 should be directories on the same fs.
1896 */
1897struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1898{
1899 struct dentry *p;
1900
1901 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001902 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 return NULL;
1904 }
1905
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001906 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001908 p = d_ancestor(p2, p1);
1909 if (p) {
1910 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1911 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1912 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 }
1914
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001915 p = d_ancestor(p1, p2);
1916 if (p) {
1917 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1918 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1919 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 }
1921
Ingo Molnarf2eace22006-07-03 00:25:05 -07001922 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1923 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 return NULL;
1925}
1926
1927void unlock_rename(struct dentry *p1, struct dentry *p2)
1928{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001929 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001931 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001932 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 }
1934}
1935
1936int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1937 struct nameidata *nd)
1938{
Miklos Szeredia95164d2008-07-30 15:08:48 +02001939 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
1941 if (error)
1942 return error;
1943
Al Viroacfa4382008-12-04 10:06:33 -05001944 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 return -EACCES; /* shouldn't it be ENOSYS? */
1946 mode &= S_IALLUGO;
1947 mode |= S_IFREG;
1948 error = security_inode_create(dir, dentry, mode);
1949 if (error)
1950 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001952 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001953 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return error;
1955}
1956
Al Viro73d049a2011-03-11 12:08:24 -05001957static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001959 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 struct inode *inode = dentry->d_inode;
1961 int error;
1962
Al Virobcda7652011-03-13 16:42:14 -04001963 /* O_PATH? */
1964 if (!acc_mode)
1965 return 0;
1966
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 if (!inode)
1968 return -ENOENT;
1969
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001970 switch (inode->i_mode & S_IFMT) {
1971 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001973 case S_IFDIR:
1974 if (acc_mode & MAY_WRITE)
1975 return -EISDIR;
1976 break;
1977 case S_IFBLK:
1978 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001979 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001981 /*FALLTHRU*/
1982 case S_IFIFO:
1983 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01001985 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08001986 }
Dave Hansenb41572e2007-10-16 23:31:14 -07001987
Christoph Hellwig3fb64192008-10-24 09:58:10 +02001988 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07001989 if (error)
1990 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 /*
1993 * An append-only file must be opened in append mode for writing.
1994 */
1995 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05001996 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05001997 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05001999 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 }
2001
2002 /* O_NOATIME can only be set by the owner or superuser */
Al Viro7715b522009-12-16 03:54:00 -05002003 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2004 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
2006 /*
2007 * Ensure there are no outstanding leases on the file.
2008 */
Al Virob65a9cf2009-12-16 06:27:40 -05002009 return break_lease(inode, flag);
Al Viro7715b522009-12-16 03:54:00 -05002010}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Jeff Laytone1181ee2010-12-07 16:19:50 -05002012static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002013{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002014 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002015 struct inode *inode = path->dentry->d_inode;
2016 int error = get_write_access(inode);
2017 if (error)
2018 return error;
2019 /*
2020 * Refuse to truncate files with mandatory locks held on them.
2021 */
2022 error = locks_verify_locked(inode);
2023 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002024 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002025 if (!error) {
2026 error = do_truncate(path->dentry, 0,
2027 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002028 filp);
Al Viro7715b522009-12-16 03:54:00 -05002029 }
2030 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002031 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032}
2033
Dave Hansend57999e2008-02-15 14:37:27 -08002034/*
Dave Hansend57999e2008-02-15 14:37:27 -08002035 * Note that while the flag value (low two bits) for sys_open means:
2036 * 00 - read-only
2037 * 01 - write-only
2038 * 10 - read-write
2039 * 11 - special
2040 * it is changed into
2041 * 00 - no permissions needed
2042 * 01 - read-permission
2043 * 10 - write-permission
2044 * 11 - read-write
2045 * for the internal routines (ie open_namei()/follow_link() etc)
2046 * This is more logical, and also allows the 00 "no perm needed"
2047 * to be used for symlinks (where the permissions are checked
2048 * later).
2049 *
2050*/
2051static inline int open_to_namei_flags(int flag)
2052{
2053 if ((flag+1) & O_ACCMODE)
2054 flag++;
2055 return flag;
2056}
2057
Nick Piggin31e6b012011-01-07 17:49:52 +11002058/*
Al Virofe2d35f2011-03-05 22:58:25 -05002059 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002060 */
Al Virofb1cc552009-12-24 01:58:28 -05002061static struct file *do_last(struct nameidata *nd, struct path *path,
Al Viroc3e380b2011-02-23 13:39:45 -05002062 const struct open_flags *op, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002063{
Al Viroa1e28032009-12-24 02:12:06 -05002064 struct dentry *dir = nd->path.dentry;
Al Viro6c0d46c2011-03-09 00:59:59 -05002065 struct dentry *dentry;
Al Viroca344a892011-03-09 00:36:45 -05002066 int open_flag = op->open_flag;
Al Viro6c0d46c2011-03-09 00:59:59 -05002067 int will_truncate = open_flag & O_TRUNC;
Al Viroca344a892011-03-09 00:36:45 -05002068 int want_write = 0;
Al Virobcda7652011-03-13 16:42:14 -04002069 int acc_mode = op->acc_mode;
Al Virofb1cc552009-12-24 01:58:28 -05002070 struct file *filp;
Al Viro16c2cd72011-02-22 15:50:10 -05002071 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002072
Al Viroc3e380b2011-02-23 13:39:45 -05002073 nd->flags &= ~LOOKUP_PARENT;
2074 nd->flags |= op->intent;
2075
Al Viro1f36f772009-12-26 10:56:19 -05002076 switch (nd->last_type) {
2077 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002078 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002079 error = handle_dots(nd, nd->last_type);
2080 if (error)
2081 return ERR_PTR(error);
Al Viro1f36f772009-12-26 10:56:19 -05002082 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002083 case LAST_ROOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002084 if (nd->flags & LOOKUP_RCU) {
2085 if (nameidata_drop_rcu_last(nd))
2086 return ERR_PTR(-ECHILD);
2087 }
Al Viro16c2cd72011-02-22 15:50:10 -05002088 error = handle_reval_path(nd);
2089 if (error)
2090 goto exit;
Al Virofe2d35f2011-03-05 22:58:25 -05002091 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002092 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002093 error = -EISDIR;
2094 goto exit;
2095 }
2096 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002097 case LAST_BIND:
Al Virofe2d35f2011-03-05 22:58:25 -05002098 /* can't be RCU mode here */
Al Viro16c2cd72011-02-22 15:50:10 -05002099 error = handle_reval_path(nd);
2100 if (error)
2101 goto exit;
Al Viro1f36f772009-12-26 10:56:19 -05002102 audit_inode(pathname, dir);
Al Viro67ee3ad2009-12-26 07:01:01 -05002103 goto ok;
Al Viro1f36f772009-12-26 10:56:19 -05002104 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002105
Al Viroca344a892011-03-09 00:36:45 -05002106 if (!(open_flag & O_CREAT)) {
Al Virobcda7652011-03-13 16:42:14 -04002107 int symlink_ok = 0;
Al Virofe2d35f2011-03-05 22:58:25 -05002108 if (nd->last.name[nd->last.len])
2109 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002110 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2111 symlink_ok = 1;
Al Virofe2d35f2011-03-05 22:58:25 -05002112 /* we _can_ be in RCU mode here */
Al Viroce57dfc2011-03-13 19:58:58 -04002113 error = walk_component(nd, path, &nd->last, LAST_NORM,
2114 !symlink_ok);
2115 if (error < 0)
Al Virofe2d35f2011-03-05 22:58:25 -05002116 return ERR_PTR(error);
Al Viroce57dfc2011-03-13 19:58:58 -04002117 if (error) /* symlink */
Al Virofe2d35f2011-03-05 22:58:25 -05002118 return NULL;
Al Virofe2d35f2011-03-05 22:58:25 -05002119 /* sayonara */
2120 if (nd->flags & LOOKUP_RCU) {
2121 if (nameidata_drop_rcu_last(nd))
2122 return ERR_PTR(-ECHILD);
2123 }
2124
2125 error = -ENOTDIR;
2126 if (nd->flags & LOOKUP_DIRECTORY) {
Al Viroce57dfc2011-03-13 19:58:58 -04002127 if (!nd->inode->i_op->lookup)
Al Virofe2d35f2011-03-05 22:58:25 -05002128 goto exit;
2129 }
2130 audit_inode(pathname, nd->path.dentry);
2131 goto ok;
2132 }
2133
2134 /* create side of things */
2135
2136 if (nd->flags & LOOKUP_RCU) {
2137 if (nameidata_drop_rcu_last(nd))
2138 return ERR_PTR(-ECHILD);
2139 }
2140
2141 audit_inode(pathname, dir);
Al Viro16c2cd72011-02-22 15:50:10 -05002142 error = -EISDIR;
Al Viro1f36f772009-12-26 10:56:19 -05002143 /* trailing slashes? */
Nick Piggin31e6b012011-01-07 17:49:52 +11002144 if (nd->last.name[nd->last.len])
2145 goto exit;
Al Viroa2c36b42009-12-24 03:39:50 -05002146
Al Viroa1e28032009-12-24 02:12:06 -05002147 mutex_lock(&dir->d_inode->i_mutex);
2148
Al Viro6c0d46c2011-03-09 00:59:59 -05002149 dentry = lookup_hash(nd);
2150 error = PTR_ERR(dentry);
2151 if (IS_ERR(dentry)) {
Al Virofb1cc552009-12-24 01:58:28 -05002152 mutex_unlock(&dir->d_inode->i_mutex);
2153 goto exit;
2154 }
2155
Al Viro6c0d46c2011-03-09 00:59:59 -05002156 path->dentry = dentry;
2157 path->mnt = nd->path.mnt;
2158
Al Virofb1cc552009-12-24 01:58:28 -05002159 /* Negative dentry, just create the file */
Al Viro6c0d46c2011-03-09 00:59:59 -05002160 if (!dentry->d_inode) {
2161 int mode = op->mode;
2162 if (!IS_POSIXACL(dir->d_inode))
2163 mode &= ~current_umask();
Al Virofb1cc552009-12-24 01:58:28 -05002164 /*
2165 * This write is needed to ensure that a
Al Viro6c0d46c2011-03-09 00:59:59 -05002166 * rw->ro transition does not occur between
Al Virofb1cc552009-12-24 01:58:28 -05002167 * the time when the file is created and when
2168 * a permanent write count is taken through
2169 * the 'struct file' in nameidata_to_filp().
2170 */
2171 error = mnt_want_write(nd->path.mnt);
2172 if (error)
2173 goto exit_mutex_unlock;
Al Viroca344a892011-03-09 00:36:45 -05002174 want_write = 1;
Al Viro9b44f1b2011-03-09 00:17:27 -05002175 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002176 open_flag &= ~O_TRUNC;
Al Viro6c0d46c2011-03-09 00:59:59 -05002177 will_truncate = 0;
Al Virobcda7652011-03-13 16:42:14 -04002178 acc_mode = MAY_OPEN;
Al Viro6c0d46c2011-03-09 00:59:59 -05002179 error = security_path_mknod(&nd->path, dentry, mode, 0);
2180 if (error)
2181 goto exit_mutex_unlock;
2182 error = vfs_create(dir->d_inode, dentry, mode, nd);
2183 if (error)
2184 goto exit_mutex_unlock;
2185 mutex_unlock(&dir->d_inode->i_mutex);
2186 dput(nd->path.dentry);
2187 nd->path.dentry = dentry;
Al Viroca344a892011-03-09 00:36:45 -05002188 goto common;
Al Virofb1cc552009-12-24 01:58:28 -05002189 }
2190
2191 /*
2192 * It already exists.
2193 */
2194 mutex_unlock(&dir->d_inode->i_mutex);
2195 audit_inode(pathname, path->dentry);
2196
2197 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002198 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002199 goto exit_dput;
2200
David Howells9875cf82011-01-14 18:45:21 +00002201 error = follow_managed(path, nd->flags);
2202 if (error < 0)
2203 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002204
2205 error = -ENOENT;
2206 if (!path->dentry->d_inode)
2207 goto exit_dput;
Al Viro9e67f362009-12-26 07:04:50 -05002208
2209 if (path->dentry->d_inode->i_op->follow_link)
Al Virofb1cc552009-12-24 01:58:28 -05002210 return NULL;
Al Virofb1cc552009-12-24 01:58:28 -05002211
2212 path_to_nameidata(path, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002213 nd->inode = path->dentry->d_inode;
Al Virofb1cc552009-12-24 01:58:28 -05002214 error = -EISDIR;
Nick Piggin31e6b012011-01-07 17:49:52 +11002215 if (S_ISDIR(nd->inode->i_mode))
Al Virofb1cc552009-12-24 01:58:28 -05002216 goto exit;
Al Viro67ee3ad2009-12-26 07:01:01 -05002217ok:
Al Viro6c0d46c2011-03-09 00:59:59 -05002218 if (!S_ISREG(nd->inode->i_mode))
2219 will_truncate = 0;
2220
Al Viro0f9d1a12011-03-09 00:13:14 -05002221 if (will_truncate) {
2222 error = mnt_want_write(nd->path.mnt);
2223 if (error)
2224 goto exit;
Al Viroca344a892011-03-09 00:36:45 -05002225 want_write = 1;
Al Viro0f9d1a12011-03-09 00:13:14 -05002226 }
Al Viroca344a892011-03-09 00:36:45 -05002227common:
Al Virobcda7652011-03-13 16:42:14 -04002228 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002229 if (error)
Al Viro0f9d1a12011-03-09 00:13:14 -05002230 goto exit;
Al Viro0f9d1a12011-03-09 00:13:14 -05002231 filp = nameidata_to_filp(nd);
2232 if (!IS_ERR(filp)) {
2233 error = ima_file_check(filp, op->acc_mode);
2234 if (error) {
2235 fput(filp);
2236 filp = ERR_PTR(error);
2237 }
2238 }
2239 if (!IS_ERR(filp)) {
2240 if (will_truncate) {
2241 error = handle_truncate(filp);
2242 if (error) {
2243 fput(filp);
2244 filp = ERR_PTR(error);
2245 }
2246 }
2247 }
Al Viroca344a892011-03-09 00:36:45 -05002248out:
2249 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002250 mnt_drop_write(nd->path.mnt);
2251 path_put(&nd->path);
Al Virofb1cc552009-12-24 01:58:28 -05002252 return filp;
2253
2254exit_mutex_unlock:
2255 mutex_unlock(&dir->d_inode->i_mutex);
2256exit_dput:
2257 path_put_conditional(path, nd);
2258exit:
Al Viroca344a892011-03-09 00:36:45 -05002259 filp = ERR_PTR(error);
2260 goto out;
Al Virofb1cc552009-12-24 01:58:28 -05002261}
2262
Al Viro13aab422011-02-23 17:54:08 -05002263static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002264 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265{
Al Virofe2d35f2011-03-05 22:58:25 -05002266 struct file *base = NULL;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002267 struct file *filp;
Al Viro9850c052010-01-13 15:01:15 -05002268 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 int count = 0;
Al Viro13aab422011-02-23 17:54:08 -05002270 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002271
2272 filp = get_empty_filp();
2273 if (!filp)
2274 return ERR_PTR(-ENFILE);
2275
Al Viro47c805d2011-02-23 17:44:09 -05002276 filp->f_flags = op->open_flag;
Al Viro73d049a2011-03-11 12:08:24 -05002277 nd->intent.open.file = filp;
2278 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2279 nd->intent.open.create_mode = op->mode;
Nick Piggin31e6b012011-01-07 17:49:52 +11002280
Al Viro73d049a2011-03-11 12:08:24 -05002281 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002282 if (unlikely(error))
Al Viro13aab422011-02-23 17:54:08 -05002283 goto out_filp;
Nick Piggin31e6b012011-01-07 17:49:52 +11002284
Al Virofe2d35f2011-03-05 22:58:25 -05002285 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002286 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002287 if (unlikely(error))
2288 goto out_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
Al Viro73d049a2011-03-11 12:08:24 -05002290 filp = do_last(nd, &path, op, pathname);
Al Viro806b6812009-12-26 07:16:40 -05002291 while (unlikely(!filp)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002292 struct path link = path;
2293 struct inode *linki = link.dentry->d_inode;
Al Virodef4af32009-12-26 08:37:05 -05002294 void *cookie;
Al Viro73d049a2011-03-11 12:08:24 -05002295 if (!(nd->flags & LOOKUP_FOLLOW) || count++ == 32) {
2296 path_put_conditional(&path, nd);
2297 path_put(&nd->path);
Al Viro40b39132011-03-09 16:22:18 -05002298 filp = ERR_PTR(-ELOOP);
2299 break;
2300 }
Al Viro806b6812009-12-26 07:16:40 -05002301 /*
2302 * This is subtle. Instead of calling do_follow_link() we do
2303 * the thing by hands. The reason is that this way we have zero
2304 * link_count and path_walk() (called from ->follow_link)
2305 * honoring LOOKUP_PARENT. After that we have the parent and
2306 * last component, i.e. we are in the same situation as after
2307 * the first path_walk(). Well, almost - if the last component
2308 * is normal we get its copy stored in nd->last.name and we will
2309 * have to putname() it when we are done. Procfs-like symlinks
2310 * just set LAST_BIND.
2311 */
Al Viro73d049a2011-03-11 12:08:24 -05002312 nd->flags |= LOOKUP_PARENT;
2313 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2314 error = __do_follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002315 if (unlikely(error))
Al Virof1afe9e2011-02-22 22:27:28 -05002316 filp = ERR_PTR(error);
Al Viroc3e380b2011-02-23 13:39:45 -05002317 else
Al Viro73d049a2011-03-11 12:08:24 -05002318 filp = do_last(nd, &path, op, pathname);
Al Virof1afe9e2011-02-22 22:27:28 -05002319 if (!IS_ERR(cookie) && linki->i_op->put_link)
Al Viro73d049a2011-03-11 12:08:24 -05002320 linki->i_op->put_link(link.dentry, nd, cookie);
Nick Piggin7b9337a2011-01-14 08:42:43 +00002321 path_put(&link);
Al Viro806b6812009-12-26 07:16:40 -05002322 }
Al Viro10fa8e62009-12-26 07:09:49 -05002323out:
Al Viro73d049a2011-03-11 12:08:24 -05002324 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2325 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002326 if (base)
2327 fput(base);
Al Viro73d049a2011-03-11 12:08:24 -05002328 release_open_intent(nd);
Al Viro10fa8e62009-12-26 07:09:49 -05002329 return filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330
Nick Piggin31e6b012011-01-07 17:49:52 +11002331out_filp:
Al Viro806b6812009-12-26 07:16:40 -05002332 filp = ERR_PTR(error);
Al Viro10fa8e62009-12-26 07:09:49 -05002333 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334}
2335
Al Viro13aab422011-02-23 17:54:08 -05002336struct file *do_filp_open(int dfd, const char *pathname,
2337 const struct open_flags *op, int flags)
2338{
Al Viro73d049a2011-03-11 12:08:24 -05002339 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002340 struct file *filp;
2341
Al Viro73d049a2011-03-11 12:08:24 -05002342 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002343 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002344 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002345 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002346 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002347 return filp;
2348}
2349
Al Viro73d049a2011-03-11 12:08:24 -05002350struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2351 const char *name, const struct open_flags *op, int flags)
2352{
2353 struct nameidata nd;
2354 struct file *file;
2355
2356 nd.root.mnt = mnt;
2357 nd.root.dentry = dentry;
2358
2359 flags |= LOOKUP_ROOT;
2360
Al Virobcda7652011-03-13 16:42:14 -04002361 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002362 return ERR_PTR(-ELOOP);
2363
2364 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2365 if (unlikely(file == ERR_PTR(-ECHILD)))
2366 file = path_openat(-1, name, &nd, op, flags);
2367 if (unlikely(file == ERR_PTR(-ESTALE)))
2368 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2369 return file;
2370}
2371
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372/**
2373 * lookup_create - lookup a dentry, creating it if it doesn't exist
2374 * @nd: nameidata info
2375 * @is_dir: directory flag
2376 *
2377 * Simple function to lookup and return a dentry and create it
2378 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002379 *
Jan Blunck4ac91372008-02-14 19:34:32 -08002380 * Returns with nd->path.dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 */
2382struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2383{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002384 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
Jan Blunck4ac91372008-02-14 19:34:32 -08002386 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002387 /*
2388 * Yucky last component or no last component at all?
2389 * (foo/., foo/.., /////)
2390 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 if (nd->last_type != LAST_NORM)
2392 goto fail;
2393 nd->flags &= ~LOOKUP_PARENT;
Al Viro35165862008-08-05 03:00:49 -04002394 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
ASANO Masahiroa6349042006-08-22 20:06:02 -04002395 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002396
2397 /*
2398 * Do the final lookup.
2399 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08002400 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 if (IS_ERR(dentry))
2402 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002403
Al Viroe9baf6e2008-05-15 04:49:12 -04002404 if (dentry->d_inode)
2405 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002406 /*
2407 * Special case - lookup gave negative, but... we had foo/bar/
2408 * From the vfs_mknod() POV we just have a negative dentry -
2409 * all is fine. Let's be bastards - you had / on the end, you've
2410 * been asking for (non-existent) directory. -ENOENT for you.
2411 */
Al Viroe9baf6e2008-05-15 04:49:12 -04002412 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2413 dput(dentry);
2414 dentry = ERR_PTR(-ENOENT);
2415 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002417eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002419 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420fail:
2421 return dentry;
2422}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07002423EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424
2425int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2426{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002427 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
2429 if (error)
2430 return error;
2431
2432 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2433 return -EPERM;
2434
Al Viroacfa4382008-12-04 10:06:33 -05002435 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 return -EPERM;
2437
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002438 error = devcgroup_inode_mknod(mode, dev);
2439 if (error)
2440 return error;
2441
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 error = security_inode_mknod(dir, dentry, mode, dev);
2443 if (error)
2444 return error;
2445
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002447 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002448 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 return error;
2450}
2451
Dave Hansen463c3192008-02-15 14:37:57 -08002452static int may_mknod(mode_t mode)
2453{
2454 switch (mode & S_IFMT) {
2455 case S_IFREG:
2456 case S_IFCHR:
2457 case S_IFBLK:
2458 case S_IFIFO:
2459 case S_IFSOCK:
2460 case 0: /* zero mode translates to S_IFREG */
2461 return 0;
2462 case S_IFDIR:
2463 return -EPERM;
2464 default:
2465 return -EINVAL;
2466 }
2467}
2468
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002469SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2470 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471{
Al Viro2ad94ae2008-07-21 09:32:51 -04002472 int error;
2473 char *tmp;
2474 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 struct nameidata nd;
2476
2477 if (S_ISDIR(mode))
2478 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
Al Viro2ad94ae2008-07-21 09:32:51 -04002480 error = user_path_parent(dfd, filename, &nd, &tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002482 return error;
2483
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 dentry = lookup_create(&nd, 0);
Dave Hansen463c3192008-02-15 14:37:57 -08002485 if (IS_ERR(dentry)) {
2486 error = PTR_ERR(dentry);
2487 goto out_unlock;
2488 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002489 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002490 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002491 error = may_mknod(mode);
2492 if (error)
2493 goto out_dput;
2494 error = mnt_want_write(nd.path.mnt);
2495 if (error)
2496 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002497 error = security_path_mknod(&nd.path, dentry, mode, dev);
2498 if (error)
2499 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002500 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 case 0: case S_IFREG:
Jan Blunck4ac91372008-02-14 19:34:32 -08002502 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 break;
2504 case S_IFCHR: case S_IFBLK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002505 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 new_decode_dev(dev));
2507 break;
2508 case S_IFIFO: case S_IFSOCK:
Jan Blunck4ac91372008-02-14 19:34:32 -08002509 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002512out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002513 mnt_drop_write(nd.path.mnt);
2514out_dput:
2515 dput(dentry);
2516out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002517 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002518 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 putname(tmp);
2520
2521 return error;
2522}
2523
Heiko Carstens3480b252009-01-14 14:14:16 +01002524SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002525{
2526 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2527}
2528
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2530{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002531 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
2533 if (error)
2534 return error;
2535
Al Viroacfa4382008-12-04 10:06:33 -05002536 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 return -EPERM;
2538
2539 mode &= (S_IRWXUGO|S_ISVTX);
2540 error = security_inode_mkdir(dir, dentry, mode);
2541 if (error)
2542 return error;
2543
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002545 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002546 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 return error;
2548}
2549
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002550SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551{
2552 int error = 0;
2553 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07002554 struct dentry *dentry;
2555 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556
Al Viro2ad94ae2008-07-21 09:32:51 -04002557 error = user_path_parent(dfd, pathname, &nd, &tmp);
2558 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002559 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Dave Hansen6902d922006-09-30 23:29:01 -07002561 dentry = lookup_create(&nd, 1);
2562 error = PTR_ERR(dentry);
2563 if (IS_ERR(dentry))
2564 goto out_unlock;
2565
Jan Blunck4ac91372008-02-14 19:34:32 -08002566 if (!IS_POSIXACL(nd.path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002567 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002568 error = mnt_want_write(nd.path.mnt);
2569 if (error)
2570 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002571 error = security_path_mkdir(&nd.path, dentry, mode);
2572 if (error)
2573 goto out_drop_write;
Jan Blunck4ac91372008-02-14 19:34:32 -08002574 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002575out_drop_write:
Dave Hansen463c3192008-02-15 14:37:57 -08002576 mnt_drop_write(nd.path.mnt);
2577out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002578 dput(dentry);
2579out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002580 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002581 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002582 putname(tmp);
2583out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 return error;
2585}
2586
Heiko Carstens3cdad422009-01-14 14:14:22 +01002587SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002588{
2589 return sys_mkdirat(AT_FDCWD, pathname, mode);
2590}
2591
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592/*
2593 * We try to drop the dentry early: we should have
2594 * a usage count of 2 if we're the only user of this
2595 * dentry, and if that is true (possibly after pruning
2596 * the dcache), then we drop the dentry now.
2597 *
2598 * A low-level filesystem can, if it choses, legally
2599 * do a
2600 *
2601 * if (!d_unhashed(dentry))
2602 * return -EBUSY;
2603 *
2604 * if it cannot handle the case of removing a directory
2605 * that is still in use by something else..
2606 */
2607void dentry_unhash(struct dentry *dentry)
2608{
2609 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002610 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002612 if (dentry->d_count == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 __d_drop(dentry);
2614 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615}
2616
2617int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2618{
2619 int error = may_delete(dir, dentry, 1);
2620
2621 if (error)
2622 return error;
2623
Al Viroacfa4382008-12-04 10:06:33 -05002624 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 return -EPERM;
2626
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002627 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 dentry_unhash(dentry);
2629 if (d_mountpoint(dentry))
2630 error = -EBUSY;
2631 else {
2632 error = security_inode_rmdir(dir, dentry);
2633 if (!error) {
2634 error = dir->i_op->rmdir(dir, dentry);
Al Virod83c49f2010-04-30 17:17:09 -04002635 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 dentry->d_inode->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04002637 dont_mount(dentry);
2638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 }
2640 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002641 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 d_delete(dentry);
2644 }
2645 dput(dentry);
2646
2647 return error;
2648}
2649
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002650static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651{
2652 int error = 0;
2653 char * name;
2654 struct dentry *dentry;
2655 struct nameidata nd;
2656
Al Viro2ad94ae2008-07-21 09:32:51 -04002657 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002659 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
2661 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002662 case LAST_DOTDOT:
2663 error = -ENOTEMPTY;
2664 goto exit1;
2665 case LAST_DOT:
2666 error = -EINVAL;
2667 goto exit1;
2668 case LAST_ROOT:
2669 error = -EBUSY;
2670 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002672
2673 nd.flags &= ~LOOKUP_PARENT;
2674
Jan Blunck4ac91372008-02-14 19:34:32 -08002675 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002676 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002678 if (IS_ERR(dentry))
2679 goto exit2;
Dave Hansen06227532008-02-15 14:37:34 -08002680 error = mnt_want_write(nd.path.mnt);
2681 if (error)
2682 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002683 error = security_path_rmdir(&nd.path, dentry);
2684 if (error)
2685 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08002686 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002687exit4:
Dave Hansen06227532008-02-15 14:37:34 -08002688 mnt_drop_write(nd.path.mnt);
2689exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07002690 dput(dentry);
2691exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08002692 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002694 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 putname(name);
2696 return error;
2697}
2698
Heiko Carstens3cdad422009-01-14 14:14:22 +01002699SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002700{
2701 return do_rmdir(AT_FDCWD, pathname);
2702}
2703
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704int vfs_unlink(struct inode *dir, struct dentry *dentry)
2705{
2706 int error = may_delete(dir, dentry, 0);
2707
2708 if (error)
2709 return error;
2710
Al Viroacfa4382008-12-04 10:06:33 -05002711 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 return -EPERM;
2713
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002714 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 if (d_mountpoint(dentry))
2716 error = -EBUSY;
2717 else {
2718 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002719 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05002721 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04002722 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05002723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002725 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726
2727 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2728 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08002729 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 }
Robert Love0eeca282005-07-12 17:06:03 -04002732
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 return error;
2734}
2735
2736/*
2737 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002738 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 * writeout happening, and we don't want to prevent access to the directory
2740 * while waiting on the I/O.
2741 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002742static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743{
Al Viro2ad94ae2008-07-21 09:32:51 -04002744 int error;
2745 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 struct dentry *dentry;
2747 struct nameidata nd;
2748 struct inode *inode = NULL;
2749
Al Viro2ad94ae2008-07-21 09:32:51 -04002750 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002752 return error;
2753
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 error = -EISDIR;
2755 if (nd.last_type != LAST_NORM)
2756 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09002757
2758 nd.flags &= ~LOOKUP_PARENT;
2759
Jan Blunck4ac91372008-02-14 19:34:32 -08002760 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002761 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 error = PTR_ERR(dentry);
2763 if (!IS_ERR(dentry)) {
2764 /* Why not before? Because we want correct error value */
2765 if (nd.last.name[nd.last.len])
2766 goto slashes;
2767 inode = dentry->d_inode;
2768 if (inode)
Al Viro7de9c6ee2010-10-23 11:11:40 -04002769 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08002770 error = mnt_want_write(nd.path.mnt);
2771 if (error)
2772 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002773 error = security_path_unlink(&nd.path, dentry);
2774 if (error)
2775 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08002776 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002777exit3:
Dave Hansen06227532008-02-15 14:37:34 -08002778 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 exit2:
2780 dput(dentry);
2781 }
Jan Blunck4ac91372008-02-14 19:34:32 -08002782 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 if (inode)
2784 iput(inode); /* truncate the inode here */
2785exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08002786 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 putname(name);
2788 return error;
2789
2790slashes:
2791 error = !dentry->d_inode ? -ENOENT :
2792 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2793 goto exit2;
2794}
2795
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002796SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002797{
2798 if ((flag & ~AT_REMOVEDIR) != 0)
2799 return -EINVAL;
2800
2801 if (flag & AT_REMOVEDIR)
2802 return do_rmdir(dfd, pathname);
2803
2804 return do_unlinkat(dfd, pathname);
2805}
2806
Heiko Carstens3480b252009-01-14 14:14:16 +01002807SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002808{
2809 return do_unlinkat(AT_FDCWD, pathname);
2810}
2811
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002812int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002814 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
2816 if (error)
2817 return error;
2818
Al Viroacfa4382008-12-04 10:06:33 -05002819 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 return -EPERM;
2821
2822 error = security_inode_symlink(dir, dentry, oldname);
2823 if (error)
2824 return error;
2825
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002827 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002828 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 return error;
2830}
2831
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002832SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2833 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834{
Al Viro2ad94ae2008-07-21 09:32:51 -04002835 int error;
2836 char *from;
2837 char *to;
Dave Hansen6902d922006-09-30 23:29:01 -07002838 struct dentry *dentry;
2839 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840
2841 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04002842 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04002844
2845 error = user_path_parent(newdfd, newname, &nd, &to);
2846 if (error)
Dave Hansen6902d922006-09-30 23:29:01 -07002847 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848
Dave Hansen6902d922006-09-30 23:29:01 -07002849 dentry = lookup_create(&nd, 0);
2850 error = PTR_ERR(dentry);
2851 if (IS_ERR(dentry))
2852 goto out_unlock;
2853
Dave Hansen75c3f292008-02-15 14:37:45 -08002854 error = mnt_want_write(nd.path.mnt);
2855 if (error)
2856 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002857 error = security_path_symlink(&nd.path, dentry, from);
2858 if (error)
2859 goto out_drop_write;
Miklos Szeredidb2e7472008-06-24 16:50:16 +02002860 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002861out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002862 mnt_drop_write(nd.path.mnt);
2863out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002864 dput(dentry);
2865out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002866 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Jan Blunck1d957f92008-02-14 19:34:35 -08002867 path_put(&nd.path);
Dave Hansen6902d922006-09-30 23:29:01 -07002868 putname(to);
2869out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 putname(from);
2871 return error;
2872}
2873
Heiko Carstens3480b252009-01-14 14:14:16 +01002874SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002875{
2876 return sys_symlinkat(oldname, AT_FDCWD, newname);
2877}
2878
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2880{
2881 struct inode *inode = old_dentry->d_inode;
2882 int error;
2883
2884 if (!inode)
2885 return -ENOENT;
2886
Miklos Szeredia95164d2008-07-30 15:08:48 +02002887 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 if (error)
2889 return error;
2890
2891 if (dir->i_sb != inode->i_sb)
2892 return -EXDEV;
2893
2894 /*
2895 * A link to an append-only or immutable file cannot be created.
2896 */
2897 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2898 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05002899 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002901 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 return -EPERM;
2903
2904 error = security_inode_link(old_dentry, dir, new_dentry);
2905 if (error)
2906 return error;
2907
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002908 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05302909 /* Make sure we don't allow creating hardlink to an unlinked file */
2910 if (inode->i_nlink == 0)
2911 error = -ENOENT;
2912 else
2913 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002914 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002915 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02002916 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 return error;
2918}
2919
2920/*
2921 * Hardlinks are often used in delicate situations. We avoid
2922 * security-related surprises by not following symlinks on the
2923 * newname. --KAB
2924 *
2925 * We don't follow them on the oldname either to be compatible
2926 * with linux 2.0, and to avoid hard-linking to directories
2927 * and other special files. --ADM
2928 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002929SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2930 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931{
2932 struct dentry *new_dentry;
Al Viro2d8f3032008-07-22 09:59:21 -04002933 struct nameidata nd;
2934 struct path old_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302935 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936 int error;
Al Viro2ad94ae2008-07-21 09:32:51 -04002937 char *to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302939 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002940 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302941 /*
2942 * To use null names we require CAP_DAC_READ_SEARCH
2943 * This ensures that not everyone will be able to create
2944 * handlink using the passed filedescriptor.
2945 */
2946 if (flags & AT_EMPTY_PATH) {
2947 if (!capable(CAP_DAC_READ_SEARCH))
2948 return -ENOENT;
2949 how = LOOKUP_EMPTY;
2950 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002951
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05302952 if (flags & AT_SYMLINK_FOLLOW)
2953 how |= LOOKUP_FOLLOW;
2954
2955 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04002957 return error;
2958
2959 error = user_path_parent(newdfd, newname, &nd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 if (error)
2961 goto out;
2962 error = -EXDEV;
Al Viro2d8f3032008-07-22 09:59:21 -04002963 if (old_path.mnt != nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 goto out_release;
2965 new_dentry = lookup_create(&nd, 0);
2966 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002967 if (IS_ERR(new_dentry))
2968 goto out_unlock;
Dave Hansen75c3f292008-02-15 14:37:45 -08002969 error = mnt_want_write(nd.path.mnt);
2970 if (error)
2971 goto out_dput;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002972 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2973 if (error)
2974 goto out_drop_write;
Al Viro2d8f3032008-07-22 09:59:21 -04002975 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002976out_drop_write:
Dave Hansen75c3f292008-02-15 14:37:45 -08002977 mnt_drop_write(nd.path.mnt);
2978out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07002979 dput(new_dentry);
2980out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -08002981 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982out_release:
Jan Blunck1d957f92008-02-14 19:34:35 -08002983 path_put(&nd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04002984 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985out:
Al Viro2d8f3032008-07-22 09:59:21 -04002986 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987
2988 return error;
2989}
2990
Heiko Carstens3480b252009-01-14 14:14:16 +01002991SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002992{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002993 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002994}
2995
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996/*
2997 * The worst of all namespace operations - renaming directory. "Perverted"
2998 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2999 * Problems:
3000 * a) we can get into loop creation. Check is done in is_subdir().
3001 * b) race potential - two innocent renames can create a loop together.
3002 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003003 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 * story.
3005 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003006 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 * whether the target exists). Solution: try to be smart with locking
3008 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003009 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 * move will be locked. Thus we can rank directories by the tree
3011 * (ancestors first) and rank all non-directories after them.
3012 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003013 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 * HOWEVER, it relies on the assumption that any object with ->lookup()
3015 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3016 * we'd better make sure that there's no link(2) for them.
3017 * d) some filesystems don't support opened-but-unlinked directories,
3018 * either because of layout or because they are not ready to deal with
3019 * all cases correctly. The latter will be fixed (taking this sort of
3020 * stuff into VFS), but the former is not going away. Solution: the same
3021 * trick as in rmdir().
3022 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003023 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003025 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 * locking].
3027 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003028static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3029 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030{
3031 int error = 0;
3032 struct inode *target;
3033
3034 /*
3035 * If we are going to change the parent - check write permissions,
3036 * we'll need to flip '..'.
3037 */
3038 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003039 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 if (error)
3041 return error;
3042 }
3043
3044 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3045 if (error)
3046 return error;
3047
3048 target = new_dentry->d_inode;
Al Virod83c49f2010-04-30 17:17:09 -04003049 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003050 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3052 error = -EBUSY;
Al Virod83c49f2010-04-30 17:17:09 -04003053 else {
3054 if (target)
3055 dentry_unhash(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 if (target) {
Al Virod83c49f2010-04-30 17:17:09 -04003059 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 target->i_flags |= S_DEAD;
Al Virod83c49f2010-04-30 17:17:09 -04003061 dont_mount(new_dentry);
3062 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003063 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 if (d_unhashed(new_dentry))
3065 d_rehash(new_dentry);
3066 dput(new_dentry);
3067 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003068 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003069 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3070 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 return error;
3072}
3073
Adrian Bunk75c96f82005-05-05 16:16:09 -07003074static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3075 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076{
3077 struct inode *target;
3078 int error;
3079
3080 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3081 if (error)
3082 return error;
3083
3084 dget(new_dentry);
3085 target = new_dentry->d_inode;
3086 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003087 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3089 error = -EBUSY;
3090 else
3091 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3092 if (!error) {
Al Virobec10522010-03-03 14:12:08 -05003093 if (target)
Al Virod83c49f2010-04-30 17:17:09 -04003094 dont_mount(new_dentry);
Mark Fasheh349457c2006-09-08 14:22:21 -07003095 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 }
3098 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003099 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 dput(new_dentry);
3101 return error;
3102}
3103
3104int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3105 struct inode *new_dir, struct dentry *new_dentry)
3106{
3107 int error;
3108 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003109 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
3111 if (old_dentry->d_inode == new_dentry->d_inode)
3112 return 0;
3113
3114 error = may_delete(old_dir, old_dentry, is_dir);
3115 if (error)
3116 return error;
3117
3118 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003119 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 else
3121 error = may_delete(new_dir, new_dentry, is_dir);
3122 if (error)
3123 return error;
3124
Al Viroacfa4382008-12-04 10:06:33 -05003125 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 return -EPERM;
3127
Robert Love0eeca282005-07-12 17:06:03 -04003128 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3129
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 if (is_dir)
3131 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3132 else
3133 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003134 if (!error)
3135 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003136 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003137 fsnotify_oldname_free(old_name);
3138
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 return error;
3140}
3141
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003142SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3143 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144{
Al Viro2ad94ae2008-07-21 09:32:51 -04003145 struct dentry *old_dir, *new_dir;
3146 struct dentry *old_dentry, *new_dentry;
3147 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003149 char *from;
3150 char *to;
3151 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152
Al Viro2ad94ae2008-07-21 09:32:51 -04003153 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 if (error)
3155 goto exit;
3156
Al Viro2ad94ae2008-07-21 09:32:51 -04003157 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 if (error)
3159 goto exit1;
3160
3161 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003162 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 goto exit2;
3164
Jan Blunck4ac91372008-02-14 19:34:32 -08003165 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 error = -EBUSY;
3167 if (oldnd.last_type != LAST_NORM)
3168 goto exit2;
3169
Jan Blunck4ac91372008-02-14 19:34:32 -08003170 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 if (newnd.last_type != LAST_NORM)
3172 goto exit2;
3173
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003174 oldnd.flags &= ~LOOKUP_PARENT;
3175 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003176 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003177
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 trap = lock_rename(new_dir, old_dir);
3179
Christoph Hellwig49705b72005-11-08 21:35:06 -08003180 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 error = PTR_ERR(old_dentry);
3182 if (IS_ERR(old_dentry))
3183 goto exit3;
3184 /* source must exist */
3185 error = -ENOENT;
3186 if (!old_dentry->d_inode)
3187 goto exit4;
3188 /* unless the source is a directory trailing slashes give -ENOTDIR */
3189 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3190 error = -ENOTDIR;
3191 if (oldnd.last.name[oldnd.last.len])
3192 goto exit4;
3193 if (newnd.last.name[newnd.last.len])
3194 goto exit4;
3195 }
3196 /* source should not be ancestor of target */
3197 error = -EINVAL;
3198 if (old_dentry == trap)
3199 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003200 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201 error = PTR_ERR(new_dentry);
3202 if (IS_ERR(new_dentry))
3203 goto exit4;
3204 /* target should not be an ancestor of source */
3205 error = -ENOTEMPTY;
3206 if (new_dentry == trap)
3207 goto exit5;
3208
Dave Hansen9079b1e2008-02-15 14:37:49 -08003209 error = mnt_want_write(oldnd.path.mnt);
3210 if (error)
3211 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003212 error = security_path_rename(&oldnd.path, old_dentry,
3213 &newnd.path, new_dentry);
3214 if (error)
3215 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 error = vfs_rename(old_dir->d_inode, old_dentry,
3217 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003218exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003219 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220exit5:
3221 dput(new_dentry);
3222exit4:
3223 dput(old_dentry);
3224exit3:
3225 unlock_rename(new_dir, old_dir);
3226exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003227 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003228 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003230 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003232exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 return error;
3234}
3235
Heiko Carstensa26eab22009-01-14 14:14:17 +01003236SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003237{
3238 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3239}
3240
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3242{
3243 int len;
3244
3245 len = PTR_ERR(link);
3246 if (IS_ERR(link))
3247 goto out;
3248
3249 len = strlen(link);
3250 if (len > (unsigned) buflen)
3251 len = buflen;
3252 if (copy_to_user(buffer, link, len))
3253 len = -EFAULT;
3254out:
3255 return len;
3256}
3257
3258/*
3259 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3260 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3261 * using) it for any given inode is up to filesystem.
3262 */
3263int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3264{
3265 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003266 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003267 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003268
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003270 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003271 if (IS_ERR(cookie))
3272 return PTR_ERR(cookie);
3273
3274 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3275 if (dentry->d_inode->i_op->put_link)
3276 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3277 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278}
3279
3280int vfs_follow_link(struct nameidata *nd, const char *link)
3281{
3282 return __vfs_follow_link(nd, link);
3283}
3284
3285/* get the link contents into pagecache */
3286static char *page_getlink(struct dentry * dentry, struct page **ppage)
3287{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003288 char *kaddr;
3289 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003291 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003293 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003295 kaddr = kmap(page);
3296 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3297 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298}
3299
3300int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3301{
3302 struct page *page = NULL;
3303 char *s = page_getlink(dentry, &page);
3304 int res = vfs_readlink(dentry,buffer,buflen,s);
3305 if (page) {
3306 kunmap(page);
3307 page_cache_release(page);
3308 }
3309 return res;
3310}
3311
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003312void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003314 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003316 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317}
3318
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003319void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003321 struct page *page = cookie;
3322
3323 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 kunmap(page);
3325 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 }
3327}
3328
Nick Piggin54566b22009-01-04 12:00:53 -08003329/*
3330 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3331 */
3332int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333{
3334 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003335 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003336 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003337 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003339 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3340 if (nofs)
3341 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342
NeilBrown7e53cac2006-03-25 03:07:57 -08003343retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003344 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003345 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003347 goto fail;
3348
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 kaddr = kmap_atomic(page, KM_USER0);
3350 memcpy(kaddr, symname, len-1);
3351 kunmap_atomic(kaddr, KM_USER0);
Nick Pigginafddba42007-10-16 01:25:01 -07003352
3353 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3354 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 if (err < 0)
3356 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003357 if (err < len-1)
3358 goto retry;
3359
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 mark_inode_dirty(inode);
3361 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362fail:
3363 return err;
3364}
3365
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003366int page_symlink(struct inode *inode, const char *symname, int len)
3367{
3368 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003369 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003370}
3371
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003372const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 .readlink = generic_readlink,
3374 .follow_link = page_follow_link_light,
3375 .put_link = page_put_link,
3376};
3377
Al Viro2d8f3032008-07-22 09:59:21 -04003378EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003379EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380EXPORT_SYMBOL(follow_down);
3381EXPORT_SYMBOL(follow_up);
3382EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3383EXPORT_SYMBOL(getname);
3384EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385EXPORT_SYMBOL(lookup_one_len);
3386EXPORT_SYMBOL(page_follow_link_light);
3387EXPORT_SYMBOL(page_put_link);
3388EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003389EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390EXPORT_SYMBOL(page_symlink);
3391EXPORT_SYMBOL(page_symlink_inode_operations);
Al Viroc9c6cac2011-02-16 15:15:47 -05003392EXPORT_SYMBOL(kern_path_parent);
Al Virod1811462008-08-02 00:49:18 -04003393EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003394EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003395EXPORT_SYMBOL(inode_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08003396EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397EXPORT_SYMBOL(unlock_rename);
3398EXPORT_SYMBOL(vfs_create);
3399EXPORT_SYMBOL(vfs_follow_link);
3400EXPORT_SYMBOL(vfs_link);
3401EXPORT_SYMBOL(vfs_mkdir);
3402EXPORT_SYMBOL(vfs_mknod);
3403EXPORT_SYMBOL(generic_permission);
3404EXPORT_SYMBOL(vfs_readlink);
3405EXPORT_SYMBOL(vfs_rename);
3406EXPORT_SYMBOL(vfs_rmdir);
3407EXPORT_SYMBOL(vfs_symlink);
3408EXPORT_SYMBOL(vfs_unlink);
3409EXPORT_SYMBOL(dentry_unhash);
3410EXPORT_SYMBOL(generic_readlink);